# watermarking

A technique for embedding detectable signals in AI-generated text to help determine whether the text originated from a language model.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## PIVOT: Distribution-Preserving Multi-BitWatermarking with Minimax-Optimal Scoring

DevFeed: [PIVOT: Distribution-Preserving Multi-BitWatermarking with Minimax-Optimal Scoring](<https://devfeed.tech/articles/pivot-distribution-preserving-multi-bitwatermarking-with-minimax-optimal-scoring-143142.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24109>)

Author: Wen Zhang, Kaiyi Pang, Minhao Bai, Shuyu Gan, Liping Zhu

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [based](<https://devfeed.tech/tags/based.md>), [class](<https://devfeed.tech/tags/class.md>), [content](<https://devfeed.tech/tags/content.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [detection](<https://devfeed.tech/tags/detection.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [effectiveness](<https://devfeed.tech/tags/effectiveness.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [free](<https://devfeed.tech/tags/free.md>), [generated-content](<https://devfeed.tech/tags/generated-content.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generator](<https://devfeed.tech/tags/generator.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [hypothesis](<https://devfeed.tech/tags/hypothesis.md>), [hypothesis-testing](<https://devfeed.tech/tags/hypothesis-testing.md>), [minimax](<https://devfeed.tech/tags/minimax.md>), [stat-ap](<https://devfeed.tech/tags/stat-ap.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

PIVOT embeds recoverable message bits while preserving each next-token probability exactly. At each generation step, a keyed pseudorandom scalar and secret vocabulary permutation determine whether probability intervals are ordered by secret rank in forward or reverse order. This creates an opposite-signed dependence between the scalar and sampled token rank for the two bit values. A pivotal rank statistic has a null distribution independent of the unknown next-token law, enabling model-free calibration and decoding. The authors frame detection and recovery as multiple-hypothesis testing, derive a moment-robust minimax-optimal score over a relaxed moment class, and establish finite-sample recovery guarantees. Experiments on two representative language models report reliable detection and recovery while preserving the output distribution.

### Source excerpt

arXiv:2611.24109v1 Announce Type: new Abstract: As LLM-generated content becomes increasingly widespread, reliable attribution calls for multi-bit watermarks that embed recoverable messages without distorting the output distribution. We introduce PIVOT, a multi-bit watermark with model-free verification based on mirrored rank-interval sampling. At each generation step, PIVOT uses a keyed pseudorandom scalar and a secret permutation of the vocabulary. The message bit determines whether next-token probability intervals are arranged in forward or reverse secret-rank order. Sampling from these mirrored layouts preserves every token probability exactly while inducing an opposite-signed dependence between the scalar and the sampled token's secret rank. This dependence yields a pivotal rank statistic whose null distribution is invariant to the unknown next-token law, enabling model-free calibration and decoding. We formulate joint watermark detection and message recovery as a multiple-hypothesis testing problem, derive a moment-robust minimax-optimal detection score over a relaxed moment class, and establish finite-sample recovery guarantees. Extensive experiments on two representative large language models validate the effectiveness of PIVOT, demonstrating reliable watermark detection and message recovery while preserving the generator's output distribution.

## Distortion-Free Semantic Watermarks for Black-Box Generative Models

DevFeed: [Distortion-Free Semantic Watermarks for Black-Box Generative Models](<https://devfeed.tech/articles/distortion-free-semantic-watermarks-for-black-box-generative-models-141505.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23208>)

Author: Aviv Yossef, Or Patashnik, Or Zamir

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Generative AI](<https://devfeed.tech/topics/generative-ai.md>), [ai-image-generator](<https://devfeed.tech/topics/ai-image-generator.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [black-box](<https://devfeed.tech/tags/black-box.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [generative](<https://devfeed.tech/tags/generative.md>), [image](<https://devfeed.tech/tags/image.md>), [images](<https://devfeed.tech/tags/images.md>), [model](<https://devfeed.tech/tags/model.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [robust](<https://devfeed.tech/tags/robust.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [watermark](<https://devfeed.tech/tags/watermark.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The proposed watermark is distortion-free in the formal sense that averaging over its secret detection key leaves the generator's output distribution unchanged. It uses only black-box sampling from the generator and black-box queries to a semantic embedding, avoiding prior approaches' reliance on autoregressive-generation or diffusion-specific access. For images, the embedding must stay stable under content-preserving transformations while distinguishing independent generations from the same prompt. The authors train an encoder for this tradeoff and report better separation than off-the-shelf alternatives, with strong aggregate detection that remains robust to various transformations. They demonstrate the scheme on open-source and closed models; the supplied abstract does not specify quantitative results or which transformations were tested.

### Source excerpt

arXiv:2611.23208v1 Announce Type: cross Abstract: We present a provably distortion-free semantic watermarking algorithm that requires only black-box sampling access to a generative model and black-box access to a semantic embedding. A watermark is called distortion-free if, averaged over its secret detection key, it leaves the generator's output distribution unchanged; this provides a formal guarantee that watermarking does not reduce output quality. Conventional such watermarks are often brittle to transformations. This is especially problematic for images, where benign operations such as compression, cropping, or resizing change the entire pixel representation. Semantic watermarks therefore require detection to be robust to content-preserving transformations. Previous methods combining these properties rely heavily on specific white-box properties of the underlying model: either to autoregressive text generation or to diffusion-specific access, such as control of the initial noise or diffusion inversion. In contrast, our construction is fully generic: It only interacts with the generative model and the semantic embedding, a function that determines the notion of "preserving content" under which detection should be robust, via sample and query access. We then instantiate this framework for image generation. We characterize the two competing properties required of the semantic embedding: it should remain stable under transformations that preserve an image's content, but should distinguish independent generations from the same prompt. We train an image encoder directly for this tradeoff. The resulting encoder improves separation while maintaining stability over off-the-shelf alternatives and enables strong aggregate detection while remaining robust to various transformations. We demonstrate the effectiveness of our watermark scheme on both open-source and closed models, which is possible only due to restriction to black-box access.

## Fundamental Limits of Statistical Watermarking for LLMs with Fixed and Optimizable Common Randomness

DevFeed: [Fundamental Limits of Statistical Watermarking for LLMs with Fixed and Optimizable Common Randomness](<https://devfeed.tech/articles/fundamental-limits-of-statistical-watermarking-for-llms-with-fixed-and-optimizable-common-randomness-139272.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19305>)

Author: Haiyun He, Yepeng Liu, Zhuoer Shen, Ziqiao Wang, Yongyi Mao, Yuheng Bu

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-it](<https://devfeed.tech/tags/cs-it.md>), [detection](<https://devfeed.tech/tags/detection.md>), [information-theory](<https://devfeed.tech/tags/information-theory.md>), [limits](<https://devfeed.tech/tags/limits.md>), [llms](<https://devfeed.tech/tags/llms.md>), [math-it](<https://devfeed.tech/tags/math-it.md>), [randomness](<https://devfeed.tech/tags/randomness.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The paper frames LLM watermarking as a statistical independence test between generated text and common randomness, jointly designing the embedder and detector to maximize detection power while controlling worst-case false positives and text distortion. It finds that optimizing common randomness does not improve fundamental performance limits over fixed randomness. Its value is instead in detector design: optimizable randomness allows a fixed, deterministic, model-agnostic detector to attain optimal performance. The paper also characterizes the performance gap for fixed uniform randomness with a fixed deterministic detector, but the supplied abstract gives no quantitative details.

### Source excerpt

arXiv:2611.19305v1 Announce Type: cross Abstract: Watermarking has emerged as an important technique for distinguishing AI-generated text from human-written text. However, most existing work is primarily empirical, relies on ad hoc watermarking designs, and lacks a fundamental understanding of the performance limits. In this paper, we propose a unified theoretical framework for statistical watermarking of large language models (LLMs), casting the problem as an independence test between generated text and common randomness and jointly designing the watermarking embedder and the detector. Our objective is to maximize detection power while controlling the worst-case false positive rate (FPR) and limiting distortion to text quality. A central focus of our framework is the design of the common randomness, which plays a critical role in both watermark embedding and detection. We characterize the fundamental limits of statistical watermarking under both fixed and optimizable common randomness. Our results show that optimizing the common randomness does not improve the fundamental limits: the best achievable performance is the same as that with fixed common randomness. Nevertheless, common randomness design remains highly important because it can substantially simplify detector design. In particular, with optimizable common randomness, a fixed, deterministic, and model-agnostic detector can achieve optimal performance. We further study a practical setting with fixed uniform common randomness and a fixed deterministic detector, and characterize its performance gap relative to the optimizable case, thereby demonstrating the practical value of optimizable common randomness.

## Detection-Efficient Watermarking for Vision--Language Agents via Risk-Constrained Strength Allocation

DevFeed: [Detection-Efficient Watermarking for Vision--Language Agents via Risk-Constrained Strength Allocation](<https://devfeed.tech/articles/detection-efficient-watermarking-for-vision-language-agents-via-risk-constrained-strength-allocation-139326.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21488>)

Author: Dumindu Samaraweera, Prashant Shekhar

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [vlm](<https://devfeed.tech/topics/vlm.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [content-authenticity](<https://devfeed.tech/tags/content-authenticity.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [detection](<https://devfeed.tech/tags/detection.md>), [language](<https://devfeed.tech/tags/language.md>), [risk](<https://devfeed.tech/tags/risk.md>), [vision](<https://devfeed.tech/tags/vision.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The framework allocates semantic watermark strength during vision-language agent decoding using predictive uncertainty and a zero-shot multimodal relevance-risk proxy. A risk penalty λ sets a continuous detection-quality trade-off at deployment without retraining. At one reported operating point, mean strength is about 59.4% lower than with fixed-strength watermarking, perplexity change falls from 0.3634 to 0.1689, and clean/paraphrased ROC-AUC is 0.9552/0.9459. At comparable mean strength, its allocation policy achieves higher ROC-AUC than fixed, entropy-only, and criticality-only policies, so the reported advantage is detection efficiency rather than superior quality. The abstract also reports substantial false-positive-rate drift from finite-sample threshold calibration and recommends reporting calibration variance.

### Source excerpt

arXiv:2611.21488v1 Announce Type: new Abstract: Vision--language agents generate descriptions, decisions, and instructions that may be reused by users, downstream agents, and automated workflows, creating a need for provenance mechanisms that support content authenticity and post-deployment auditing without unnecessarily degrading response quality. We present a risk-aware semantic watermarking framework that combines predictive uncertainty with a zero-shot multimodal relevance-risk proxy to allocate watermark strength during decoding. A single risk penalty $\lambda$ exposes a continuous operating frontier, allowing deployment-time selection of a detection--quality trade-off without model retraining. At a selected operating point, mean watermark strength is reduced by approximately 59.4\% relative to fixed-strength watermarking, while perplexity change falls from 0.3634 to 0.1689 and clean/paraphrased ROC-AUC reaches 0.9552/0.9459. A matched-strength analysis shows that this advantage reflects the allocation policy rather than lower average intervention: at comparable mean strength, risk-constrained allocation achieves substantially higher ROC-AUC than fixed, entropy-only, and criticality-only policies, identifying it as detection-efficient rather than quality-dominant. We further show that threshold calibration on finite samples produces substantial false-positive-rate drift, motivating calibration variance reporting as standard practice for auditable provenance.

## Beyond Semantic Prompts: Unified Watermarking and Access Control for Text-to-Image Diffusion Models

DevFeed: [Beyond Semantic Prompts: Unified Watermarking and Access Control for Text-to-Image Diffusion Models](<https://devfeed.tech/articles/beyond-semantic-prompts-unified-watermarking-and-access-control-for-text-to-image-diffusion-models-139302.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20470>)

Author: Fangming Shi, Li Li, Guorui Feng, Xinpeng Zhang

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [text-to-image](<https://devfeed.tech/topics/text-to-image.md>), [ai-image-generator](<https://devfeed.tech/topics/ai-image-generator.md>), [Access Control](<https://devfeed.tech/topics/access-control.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [access-control](<https://devfeed.tech/tags/access-control.md>), [autoencoder](<https://devfeed.tech/tags/autoencoder.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [customizable](<https://devfeed.tech/tags/customizable.md>), [decoder](<https://devfeed.tech/tags/decoder.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [generation](<https://devfeed.tech/tags/generation.md>), [image-processing](<https://devfeed.tech/tags/image-processing.md>), [images](<https://devfeed.tech/tags/images.md>), [models](<https://devfeed.tech/tags/models.md>), [prompt-structure](<https://devfeed.tech/tags/prompt-structure.md>), [prompts](<https://devfeed.tech/tags/prompts.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [quality](<https://devfeed.tech/tags/quality.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [services](<https://devfeed.tech/tags/services.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The proposed framework combines multi-bit watermarking and tiered access control in a single text prompt. It encodes watermark messages as designated words with dedicated learnable tokens, then jointly adapts the diffusion U-Net and token embeddings using a frozen reference model and pretrained watermark codec. A prompt-structure recognition network controls the VAE decoder to produce clear watermarked images, blurred previews, or black outputs according to access state. The abstract reports reliable watermark recovery, robustness to tested image-processing operations and model modifications, and preserved decoding fidelity and watermark recovery under full access. It does not provide quantitative results in the supplied excerpt.

### Source excerpt

arXiv:2611.20470v1 Announce Type: new Abstract: Text-to-image diffusion services require both traceability of generated content and differentiated access to generation capabilities. However, existing watermarking methods commonly expose watermark control through mechanisms beyond the text-prompt interface, while credential-based access control does not inherently enforce distinct output quality levels. We propose a unified framework that enables customizable multi-bit watermark embedding and tiered access control through a single text prompt. A unified prompt represents watermark messages using designated words and associates protocol structures with different access states. The framework adopts a two-stage training strategy. First, watermark payload partitioning represents messages as designated watermark words, while vocabulary extension assigns each protocol word a dedicated token with a learnable input embedding. Guided by a frozen reference model and a pretrained watermark codec, joint adaptation of the U-Net and protocol-token embeddings transfers message-dependent watermark patterns into the generation process. Second, a prompt-structure recognition network controls feature modulation within the variational autoencoder decoder, which is trained to produce clear watermarked images, blurred previews, or black outputs according to the access state. Experiments demonstrate reliable watermark recovery, robustness to common image-processing operations and evaluated model modifications, and consistent performance across diverse generation settings. The controlled decoder also achieves distinct output behaviors under restricted access while closely preserving reference decoding fidelity and watermark recoverability under full access. Our open-source code and model weights will be public after acceptance.

## SAMark: A Self-Anchored Text Watermarking with Paragraph-Level Paraphrase Robustness

DevFeed: [SAMark: A Self-Anchored Text Watermarking with Paragraph-Level Paraphrase Robustness](<https://devfeed.tech/articles/samark-a-self-anchored-text-watermarking-with-paragraph-level-paraphrase-robustness-139232.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.25796>)

Author: Jiahao Huo, Wenjie Qu, Yibo Yan, Kening Zheng, Jiaheng Zhang, Xuming Hu, Philip S. Yu, Mingxun Zhou

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Generative AI](<https://devfeed.tech/topics/generative-ai.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [quality](<https://devfeed.tech/tags/quality.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [text](<https://devfeed.tech/tags/text.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

SAMark targets paragraph-level paraphrasing, which can disrupt sentence-based watermark signals by reordering sentences. It uses a step-independent green region in semantic space to avoid relying on sentence order, multi-channel hyperbolic scoring to strengthen detection while reducing noise from weakly aligned candidates, and diversity-aware filtering to address semantic redundancy through hard filtering and soft regularization. The abstract reports up to 90.2% TP@FP1% under typical paragraph-level paraphrasing attacks and more than 30% average improvement over the strongest prior baseline, with generation quality competitive with unwatermarked text. These are abstract-reported results; the supplied excerpt does not include experimental details.

### Source excerpt

arXiv:2605.25796v4 Announce Type: replace Abstract: Semantic-level watermarking (SWM) improves robustness against text modifications by treating sentences as the basic unit. However, robustness to paragraph-level paraphrasing remains difficult because such attacks globally disrupt watermark signals by changing sentence order. In this work, we propose SAMark, a self-anchored watermarking framework that removes the dependency on sentence order by establishing a step-independent green region in semantic space. To improve detectability, we introduce a multi-channel hyperbolic scoring mechanism that amplifies watermark signals while suppressing noise from weakly aligned candidates. We further propose a diversity-aware filtering strategy that combines hard filtering with soft regularization, extending beyond simple n-gram repetition filters to address semantic redundancy. Experimental results show that SAMark achieves up to 90.2% TP@FP1% under typical paragraph-level paraphrasing attacks, outperforming the strongest prior baseline by more than 30% on average, while maintaining generation quality competitive with unwatermarked text and breaking the robustness-quality trade-off that limits prior methods. Our code will be released at [this URL](https://github.com/Z1zs/SAMark).

## Look It Up: Multi-bit LLM Watermarking via Codebook Search

DevFeed: [Look It Up: Multi-bit LLM Watermarking via Codebook Search](<https://devfeed.tech/articles/look-it-up-multi-bit-llm-watermarking-via-codebook-search-132289.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16890>)

Author: Jaros{\\l}aw Janas, Pawe{\\l} Morawiecki, Josef Pieprzyk

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [attribution](<https://devfeed.tech/tags/attribution.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [generation](<https://devfeed.tech/tags/generation.md>), [hamming](<https://devfeed.tech/tags/hamming.md>), [levenshtein](<https://devfeed.tech/tags/levenshtein.md>), [levenshtein-distance](<https://devfeed.tech/tags/levenshtein-distance.md>), [llm](<https://devfeed.tech/tags/llm.md>), [models](<https://devfeed.tech/tags/models.md>), [payload](<https://devfeed.tech/tags/payload.md>), [probability](<https://devfeed.tech/tags/probability.md>), [quality](<https://devfeed.tech/tags/quality.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [resilience](<https://devfeed.tech/tags/resilience.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [search](<https://devfeed.tech/tags/search.md>), [storage](<https://devfeed.tech/tags/storage.md>), [tokenization](<https://devfeed.tech/tags/tokenization.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The method encodes an n-bit payload as a longer codeword, embeds it through token-level logit biasing, and extracts it by searching a secret codebook for the nearest match using Hamming or Levenshtein distance. The authors report near-perfect recovery of 16-24-bit payloads in 200-token generations and strong results at 127 tokens for base models. Recovery is weaker for some instruction-tuned models on very short texts, where low-entropy outputs limit embedding reliability. Levenshtein matching better handles deletions and light paraphrasing, but insertions and stronger rewrites remain challenging. A BCH variant avoids storing a codebook but generally reduces robustness.

### Source excerpt

arXiv:2611.16890v1 Announce Type: cross Abstract: Attributing short LLM-generated texts to a specific source remains challenging for existing multi-bit watermarking methods, which often require long generations or costly decoding to recover payloads reliably. We propose a codebook-based watermarking scheme that maps an $n$-bit payload to a longer $u$-bit codeword and embeds it during generation through token-level logit biasing. During extraction, the received bit sequence is matched to the nearest codeword in a precomputed secret codebook using Hamming or Levenshtein distance. Experiments show that our method achieves near-perfect recovery for 16-24 bit payloads on 200-token generations and strong performance on 127-token generations for base models. Results are weaker on some instruction-tuned models with very short texts, where low-entropy output distributions reduce embedding reliability. We also study a BCH-based variant that avoids codebook storage but generally reduces robustness. The Levenshtein variant improves resilience to deletions and light paraphrasing, while insertions and stronger rewrites remain challenging. We analyze collision probability, codeword separation, extraction time, tokenization consistency, and text quality, highlighting both the practicality and limitations of codebook-based watermarking for short-form attribution.

## Prevalence Bounds for Edited Watermarked Text: Power Envelopes and Calibration Sensitivity

DevFeed: [Prevalence Bounds for Edited Watermarked Text: Power Envelopes and Calibration Sensitivity](<https://devfeed.tech/articles/prevalence-bounds-for-edited-watermarked-text-power-envelopes-and-calibration-sensitivity-133166.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17035>)

Author: Kuancheng Ye

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [alternatives](<https://devfeed.tech/tags/alternatives.md>), [budget](<https://devfeed.tech/tags/budget.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [coupling](<https://devfeed.tech/tags/coupling.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [edits](<https://devfeed.tech/tags/edits.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [error](<https://devfeed.tech/tags/error.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [gap](<https://devfeed.tech/tags/gap.md>), [green](<https://devfeed.tech/tags/green.md>), [independent](<https://devfeed.tech/tags/independent.md>), [keys](<https://devfeed.tech/tags/keys.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [list](<https://devfeed.tech/tags/list.md>), [power](<https://devfeed.tech/tags/power.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [text](<https://devfeed.tech/tags/text.md>)

### AI overview

The paper bounds the share of a corpus produced by watermarked models despite unknown edits, using a power envelope for Gumbel-max and KGW watermarks. The envelope depends on a retention floor, an entropy budget, and a substitution budget; without edit restrictions, the upper bound is 1. It yields finite-sample bounds for fixed corpora and confidence sequences for streams, using either an exact null or one calibrated on independent human data. In reported experiments, 12,000 fresh human documents showed no robust departure from the exact null; at a watermark share of 0.2, the exact-null bound was about [0.14, 0.38]. Coverage was fragile to null misspecification at large sample sizes: simulated error of 0.01 broke exact-null coverage at 100,000 documents, while independent-human calibration retained coverage but required a large reference sample to stay narrow. LLM-paraphrased KGW pools fell below the envelope, though coverage held in the reported 2,000-document trials.

### Source excerpt

arXiv:2611.17035v1 Announce Type: cross Abstract: Watermark detectors decide whether one document came from a watermarked language model. Auditors often need a different number: the share of a corpus that did. We study this share for Gumbel-max and green-list (KGW) watermarks when every document may have been edited by an unknown amount. Without restrictions on the edits, the upper end of the identified set is 1. Combining a known least-favourable next-token distribution with a sequential coupling, we derive a power envelope indexed by three budgets: a retention floor, an entropy budget, and a budget for in-place substitutions, which for Gumbel-max make edited positions look less watermarked than human text. The envelope gives an identified set that is sharp for the alternatives it dominates, finite-sample bounds on the realized share of a fixed corpus with either the exact null or a null calibrated on an independent human sample, and confidence sequences for streams. In Gumbel-max simulations where the envelope binds, the bounds cover, and the threshold that maximizes the gap between envelope and null is generally as tight as twelve thresholds. On text from OPT-1.3B and Qwen3-1.7B, 12,000 fresh human documents scored under three keys show no robust departure from the exact null, and all bounds covered in resampling trials from finite pools of documents with random edits, substitutions and paraphrases; at a share of 0.2 the exact-null bound is about [0.14,0.38]. These results are fragile in two ways we make precise. LLM-paraphrased KGW pools fall below the envelope, yet at 2000 documents the bounds still covered in these trials. A simulated null error of 0.01, comparable to the largest deviation we observed at our thresholds, is likewise harmless at 2000 documents but breaks exact-null coverage at 100,000; a null calibrated on an independent human sample keeps coverage there, but needs a large reference sample to stay narrow.

## Auditing Cross-Lingual Fairness in Language Model Watermarking

DevFeed: [Auditing Cross-Lingual Fairness in Language Model Watermarking](<https://devfeed.tech/articles/auditing-cross-lingual-fairness-in-language-model-watermarking-128567.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.20047>)

Author: Alexander Nemecek, Osama Zafar, Debargha Ganguly, Vikash Singh, Vipin Chaudhary, Erman Ayday

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [auditing](<https://devfeed.tech/tags/auditing.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fairness](<https://devfeed.tech/tags/fairness.md>), [framework](<https://devfeed.tech/tags/framework.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [languages](<https://devfeed.tech/tags/languages.md>), [model](<https://devfeed.tech/tags/model.md>), [quality](<https://devfeed.tech/tags/quality.md>), [scripts](<https://devfeed.tech/tags/scripts.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The paper proposes a multilingual fairness audit for language-model watermarking that calibrates detection thresholds to each deployment context and adds a threshold-independent measure to distinguish calibration problems from detection failures. It assesses output quality through three separate paradigms--distributional, paired-semantic, and reference-perplexity--and decomposes cross-language disparities across typological families. In tests of six watermarking schemes with three open-weight generators across eleven languages, four scripts, eight typological families, and base and instruction-tuned regimes, detection disparities were predominantly between families against a shape-matched null. Quality disparities were large but did not reliably follow family boundaries. The abstract argues these patterns can be missed by English-only or single-paradigm evaluations.

### Source excerpt

arXiv:2608.20047v2 Announce Type: replace-cross Abstract: Watermarking schemes for large language model output are evaluated almost exclusively on English text using each scheme's detection threshold and a narrow set of quality measurements. Multilingual deployment exposes evaluation-design choices that are inconsequential on English but determine conclusions cross-lingually. We propose an evaluation framework with four components: detection thresholds calibrated empirically per deployment context, a threshold-independent companion measurement that distinguishes calibration failures from detection failures, three disjoint quality measurement paradigms (distributional, paired-semantic, and reference-perplexity), and a generalized-entropy decomposition of cross-language disparity over a typological family partition. Applied to six watermarking schemes, three open-weight generators, eleven languages spanning four scripts and eight typological families, and both base and instruction-tuned regimes, the framework reveals failure modes that single-language, single-paradigm evaluation cannot surface. Against a shape-matched null, detection disparity is predominantly between-family on the typological partition, indicating that detection gaps are structural to language properties rather than idiosyncratic to particular languages; quality disparity is large but not reliably family-structured.

## SpaceXSI quietly flips Grok API text watermarking to off by default

DevFeed: [SpaceXSI quietly flips Grok API text watermarking to off by default](<https://devfeed.tech/articles/spacexsi-quietly-flips-grok-api-text-watermarking-to-off-by-default-126608.md>)

Original publisher: [Read original article](<https://thenewstack.io/grok-api-watermarking-default/>)

Author: Paul Sawers

Published: 2026-11-10T15:58:44Z

Content type: news

Language: en

Sources: [The New Stack](<https://devfeed.tech/sources/the-new-stack.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [synthid](<https://devfeed.tech/topics/synthid.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [about](<https://devfeed.tech/tags/about.md>), [act](<https://devfeed.tech/tags/act.md>), [administration](<https://devfeed.tech/tags/administration.md>), [administrator](<https://devfeed.tech/tags/administrator.md>), [after](<https://devfeed.tech/tags/after.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-companies](<https://devfeed.tech/tags/ai-companies.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [api](<https://devfeed.tech/tags/api.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [first](<https://devfeed.tech/tags/first.md>), [grok](<https://devfeed.tech/tags/grok.md>), [issue](<https://devfeed.tech/tags/issue.md>), [spacexsi](<https://devfeed.tech/tags/spacexsi.md>), [stack](<https://devfeed.tech/tags/stack.md>), [text](<https://devfeed.tech/tags/text.md>), [this](<https://devfeed.tech/tags/this.md>), [turning](<https://devfeed.tech/tags/turning.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

SpaceXSI's revised documentation makes Grok API text watermarking opt-in: it is off by default across seven models, including Grok 4.7 and 4.6. A team administrator can enable it per model, and the setting applies to every API key on that team; developers cannot toggle it per request, so serving different regional policies may require separate team configurations. The feature uses Google's SynthID Text. The article says earlier documentation had scheduled default-on watermarking from November 7, but the change to opt-in occurred around November 9-10; it is unclear whether the default-on policy ever took effect or why SpaceXSI reversed course. The EU AI Act's machine-readable marking provisions are described as taking effect August 2, 2026, with a limited grace period to December 2 for qualifying systems already on the market. The article contrasts Anthropic's global, non-optional approach on supported Claude products, OpenAI's optional API watermarking and automatic watermarking for eligible EU ChatGPT and Codex output, and Google's use of SynthID across products including Gemini API.

### Source excerpt

The thorny issue of watermarking AI-generated text took an interesting turn this past week, with SpaceXSI seemingly U-turning on plans The post SpaceXSI quietly flips Grok API text watermarking to off by default appeared first on The New Stack.

## How Short Is Too Short? Length-Conditional Calibration and Abstention for Detecting AI-Generated Text

DevFeed: [How Short Is Too Short? Length-Conditional Calibration and Abstention for Detecting AI-Generated Text](<https://devfeed.tech/articles/how-short-is-too-short-length-conditional-calibration-and-abstention-for-detecting-ai-generated-text-126041.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10901>)

Author: Ritesh Janga, Justin Zhan

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [data](<https://devfeed.tech/tags/data.md>), [detection](<https://devfeed.tech/tags/detection.md>), [false-positive-rate](<https://devfeed.tech/tags/false-positive-rate.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [text](<https://devfeed.tech/tags/text.md>), [token](<https://devfeed.tech/tags/token.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

AI-text detector reliability varies sharply with passage length, so thresholds calibrated on longer passages can misstate false-positive rates on short ones. The study derives formulas for this length effect and for the tokens needed to reach a target true-positive rate, then calibrates thresholds by length bin using conformal p-values and abstains where detection is insufficiently informative. Across three zero-shot detectors, three corpora and two open-weight generators, bin-wise calibration averaged a 5.0% false-positive rate in every detector-length cell from 10 to 200 tokens. A single threshold set at 200 tokens instead produced 29% false positives for Binoculars at 10 tokens, while log-likelihood was about 1%. No detector reached an 80% true-positive rate; Binoculars peaked at 59% at 200 tokens. The abstention rule therefore withheld all cells at an 80% floor, releasing only 200-token bins at lower floors of 0.3 and 0.5.

### Source excerpt

arXiv:2611.10901v1 Announce Type: new Abstract: Detectors of AI-generated text are usually tuned and tested on passages of several hundred tokens, yet many real cases are short. We study score-based detection as a function of passage length. We give a closed form for how the false-positive rate of a threshold set on long text changes as length falls, and a formula for the number of tokens a detector needs to reach a target true-positive rate. We calibrate thresholds within length bins with conformal p-values, and we add a bin-level rule that abstains when the detector cannot be informative at that length. On synthetic scores the formulas agree with simulation, and binned calibration holds the false-positive rate at its nominal value at every length. On three zero-shot detectors (log-likelihood, log-rank, Binoculars), three corpora and two open-weight generators, binned calibration gives a mean false-positive rate of 5.0% in every detector-length cell from 10 to 200 tokens, averaged over 300 random document-level resplits. A single threshold set at 200 tokens gives 29% for Binoculars at 10 tokens and about 1% for log-likelihood, as the closed form predicts. No detector reaches an 80% true-positive rate, and the best, Binoculars at 200 tokens, reaches 59%, so the rule abstains in every cell at that floor and releases only the 200-token bins at floors of 0.3 and 0.5.

## Ultrasound-Assisted Biometric Watermarking Against Face Video Deepfakes

DevFeed: [Ultrasound-Assisted Biometric Watermarking Against Face Video Deepfakes](<https://devfeed.tech/articles/ultrasound-assisted-biometric-watermarking-against-face-video-deepfakes-126181.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11800>)

Author: Guodong Liu, Lei Wang, Dongxu Liu, Qianran Qiao, Meng Li, Haipeng Dai, He Huang, Guihai Chen

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Facial recognition](<https://devfeed.tech/topics/facial-recognition.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [breaches](<https://devfeed.tech/tags/breaches.md>), [built-in](<https://devfeed.tech/tags/built-in.md>), [capture](<https://devfeed.tech/tags/capture.md>), [content](<https://devfeed.tech/tags/content.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [deepfakes](<https://devfeed.tech/tags/deepfakes.md>), [devices](<https://devfeed.tech/tags/devices.md>), [experimental](<https://devfeed.tech/tags/experimental.md>), [forgery](<https://devfeed.tech/tags/forgery.md>), [generated-content](<https://devfeed.tech/tags/generated-content.md>), [microphone](<https://devfeed.tech/tags/microphone.md>), [mobile-devices](<https://devfeed.tech/tags/mobile-devices.md>), [offline](<https://devfeed.tech/tags/offline.md>), [online](<https://devfeed.tech/tags/online.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [protection](<https://devfeed.tech/tags/protection.md>), [ultrasound](<https://devfeed.tech/tags/ultrasound.md>), [video](<https://devfeed.tech/tags/video.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

UltraGuard embeds real-time, content-aware watermarks in facial videos by combining biometric features captured through a smartphone's built-in speaker and microphone emitting ultrasonic signals with spatio-temporal features from the video. The authors report experiments showing defense against forgery and editing at the source and during propagation. They present deployment on users' mobile devices as a future possibility; the supplied abstract does not provide implementation details or quantitative results.

### Source excerpt

arXiv:2611.11800v1 Announce Type: cross Abstract: Currently, artificial intelligence-generated content (AIGC) poses a severe threat to facial video deepfakes, and many AIGC companies employ video watermarking to mitigate unauthorized fake video usage, but its effectiveness is limited due to watermark tampering and private-model forgery. Additionally, some schemes actively add watermarks to user videos. However, current approaches focus primarily on the quality of watermark embedding in offline scenarios, and no existing work offers a solution for protecting human facial videos in real-time online applications. To fill this gap, we propose, to the best of our knowledge, a novel biometric video watermarking method, UltraGuard, for the first time. It uses the smartphone's built-in speaker and microphone to emit ultrasonic signals and capture biometric features, which are then fused with spatio-temporal dynamic features derived from the video content. The fused features are encoded as watermarks and embedded into videos in real time, enabling dynamic, content-aware watermarking throughout transmission and propagation. Extensive experimental validation demonstrates that UltraGuard can effectively defend against malicious attacks such as forgery and editing, both at the source and during propagation, proving the system's efficacy. In the future, it can be deployed on users' mobile devices to provide real-time protection for users' facial videos, thereby preventing security and privacy breaches caused by various malicious attacks.

## OpenMarkX: An Extensible Platform for Watermarking Open-Source Model and Agent Outputs

DevFeed: [OpenMarkX: An Extensible Platform for Watermarking Open-Source Model and Agent Outputs](<https://devfeed.tech/articles/openmarkx-an-extensible-platform-for-watermarking-open-source-model-and-agent-outputs-126873.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2510.23891>)

Author: Jiaqi Xue, Yifei Zhao, Mansour Al Ghanim, Shangqian Gao, Ruimin Sun, Qian Lou, Mengxin Zheng

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [after](<https://devfeed.tech/tags/after.md>), [agent](<https://devfeed.tech/tags/agent.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [field](<https://devfeed.tech/tags/field.md>), [image](<https://devfeed.tech/tags/image.md>), [inference](<https://devfeed.tech/tags/inference.md>), [keys](<https://devfeed.tech/tags/keys.md>), [language](<https://devfeed.tech/tags/language.md>), [learning](<https://devfeed.tech/tags/learning.md>), [llms](<https://devfeed.tech/tags/llms.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [platform](<https://devfeed.tech/tags/platform.md>), [source](<https://devfeed.tech/tags/source.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

OpenMarkX packages watermarking for open-source model and agent outputs under standard inference: given models, compatible methods, private keys, and evaluation settings, it produces watermarked weights, detection results, and quality and robustness reports. It provides 16 text-watermark configurations plus agent, vision-language, image, and physics-field workflows. Its proposed OpenMark method uses joint model-policy learning and perturbation-aware training to address learnability and robustness. On several open-source LLMs, it reports 0.92 TPR at 1% FPR and 20.5% lower perplexity on average than the strongest prior method. After model modifications, it remained more detectable; for example, at 5% FPR after 0.5-ratio model merging, it retained 0.57 TPR, versus at most 0.34 for prior methods.

### Source excerpt

arXiv:2510.23891v2 Announce Type: replace Abstract: We introduce OpenMarkX, an extensible platform for watermarking open-source model and agent outputs under standard inference. It takes models, compatible methods, private keys, and evaluation settings, and produces watermarked weights, detection results, and quality and robustness reports. Its 16 text-watermark configurations are complemented by agent, vision-language, image, and physics-field workflows. Our proposed method, OpenMark, addresses learnability and robustness through joint model--policy learning and perturbation-aware training. On several open-source LLMs, OpenMark achieves 0.92 TPR at 1% FPR while reducing perplexity by 20.5% on average relative to the strongest prior method, and remains more detectable after such modifications, e.g., retaining 0.57 TPR at 5% FPR after 0.5-ratio model merging, compared with at most 0.34 for prior methods.

## SkillCODER: Towards Auditing and Attribution of Copyright Infringement in LLM Agent Skills

DevFeed: [SkillCODER: Towards Auditing and Attribution of Copyright Infringement in LLM Agent Skills](<https://devfeed.tech/articles/skillcoder-towards-auditing-and-attribution-of-copyright-infringement-in-llm-agent-skills-126952.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11308>)

Author: Enhao Huang, Chunshu Xia, Yiming Li, Yuchen Yang, Bingrun Yang, Zhan Qin, Dacheng Tao, Kui Ren

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [agent-frameworks](<https://devfeed.tech/tags/agent-frameworks.md>), [agent-skills](<https://devfeed.tech/tags/agent-skills.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [compression](<https://devfeed.tech/tags/compression.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [content](<https://devfeed.tech/tags/content.md>), [copyright](<https://devfeed.tech/tags/copyright.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [domains](<https://devfeed.tech/tags/domains.md>), [embedded](<https://devfeed.tech/tags/embedded.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [error-correcting-codes](<https://devfeed.tech/tags/error-correcting-codes.md>), [instructions](<https://devfeed.tech/tags/instructions.md>), [language](<https://devfeed.tech/tags/language.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [models](<https://devfeed.tech/tags/models.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

SkillCODER embeds copyright and buyer fingerprints in LLM agent skills while aiming to preserve their task fidelity and readability. It parses each skill into a section-aware SkillIR, distributes conditional instructions across structural anchors in an Artifact-Gated Capsule, and encodes buyer identity through controlled lexical choices and error-correcting codes. Verification uses paired positive and negative probes to recover ownership and buyer identity. In evaluations across three domains, six models, and three agent frameworks, it reports 99.2% owner verification and 99.3% buyer Top-1 attribution, outperforming prompt-level baselines. Signals remained robust to paraphrase, compression, clause deletion, and reorganization. Under five adaptive attacks, ownership evidence remained reliably detectable, while buyer attribution degraded gracefully.

### Source excerpt

arXiv:2611.11308v1 Announce Type: new Abstract: Agent skills package reusable workflows and task-specific procedures into deployable natural-language artifacts for LLM agents. Their scale and readability make them valuable but vulnerable to unauthorized redistribution and repackaging. Existing prompt copyright auditing methods struggle with long skills and introduce token-level perturbations that compromise readability. We propose SkillCODER for copyright auditing and buyer attribution in a black-box, content-only setting. During embedding, SkillCODER parses a skill into a section-aware intermediate representation, SkillIR, to identify structural anchors. It distributes conditional instructions across these anchors to form an Artifact-Gated Capsule, designed to activate under verification queries and remain dormant otherwise. Controlled-Vocabulary Error-Correcting Codes encode buyer-specific fingerprints through controlled lexical choices within the capsule. During verification, the defender issues paired positive and negative probes, extracts capsule evidence, and recovers ownership and buyer identity through error-correcting decoding. Across three domains, six models, and three agent frameworks, SkillCODER achieves 99.2% owner verification and 99.3% buyer Top-1 attribution accuracy while preserving task fidelity, substantially outperforming prompt-level baselines. Embedded signals remain robust under paraphrase, compression, clause deletion, and reorganization. Under five adaptive attacks, including a method-aware LLM sanitizer and multi-buyer collusion, ownership evidence remains reliably detectable while buyer attribution degrades gracefully.

## NotaryMark: Multi-Bit Watermarking for Black-Box DatasetUsage Auditing in Text-to-Image Diffusion Models

DevFeed: [NotaryMark: Multi-Bit Watermarking for Black-Box DatasetUsage Auditing in Text-to-Image Diffusion Models](<https://devfeed.tech/articles/notarymark-multi-bit-watermarking-for-black-box-datasetusage-auditing-in-text-to-image-diffusion-models-122794.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08985>)

Author: Xinyi Yu, Linkang Du, Zhou Su, Min Chen, Yuntao Wang, Zhikun Zhang

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [Cryptography](<https://devfeed.tech/topics/cryptography.md>), [AI software supply chain security](<https://devfeed.tech/topics/ai-software-supply-chain-security.md>)

Tags: [aggregation](<https://devfeed.tech/tags/aggregation.md>), [ai](<https://devfeed.tech/tags/ai.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [bits](<https://devfeed.tech/tags/bits.md>), [budget](<https://devfeed.tech/tags/budget.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [copyright](<https://devfeed.tech/tags/copyright.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [error-correcting-codes](<https://devfeed.tech/tags/error-correcting-codes.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

NotaryMark aligns multi-bit watermarks with diffusion models' latent denoising to reduce the signal loss that limits earlier watermark methods in text-to-image training. Across three datasets and three models, reported average bit accuracy is 0.85-0.90 below 32-bit capacity and 0.67-0.80 above it, versus about 0.42-0.60 for existing multi-bit methods. Voting improves accuracy using 30 queries, and error-correcting codes support decoding and tracing embedded identifiers. Tests of 15 evasion attacks across seven categories found robustness in most evaluated settings; stronger attacks generally suppressed watermark evidence while substantially degrading model utility.

### Source excerpt

arXiv:2611.08985v1 Announce Type: new Abstract: As generative artificial intelligence (AI) models increasingly require data, more publicly available yet licensed data is being harvested and used without authorization. Although several previous studies have proposed watermarking methods for dataset copyright auditing, their applicability to the text-to-image (T2I) setting remains notably limited. The core drawback is that watermarks embedded by existing techniques are often treated as noise and are gradually filtered out and attenuated during the latent noising-denoising process of diffusion model training. We propose NotaryMark, to our knowledge the first multi-bit watermarking method specifically designed for black-box auditing of T2I training dataset use. The core idea of NotaryMark is to align the watermark with the latent denoising of diffusion models, thereby overcoming the signal decay that plagues existing multi-bit methods during sampling. We evaluate NotaryMark across three datasets and three diffusion models spanning watermark capacities from 8 to 1024 bits. With a watermark capacity below 32 bits, it achieves a high average bit accuracy of 0.85-0.90. With a watermark capacity exceeding 32 bits, it maintains bit accuracy between 0.67 and 0.80, consistently outperforming existing multi-bit watermark methods whose bit accuracy is around 0.42-0.60. Furthermore, we design a voting-based aggregation scheme to boost the bit accuracy with a modest budget of 30 queries. Building on this, we incorporate error-correcting codes to enable reliable decoding and tracing of the embedded identifier information. Robustness evaluation against fifteen evasion attacks spanning seven categories shows that NotaryMark remains robust under most evaluated settings, while stronger attacks generally suppress watermark evidence only at the cost of substantial utility degradation, demonstrating its practical reliability against malicious tampering.

## Logit Laundering: Evading Data-Use Auditing in Large Language Models

DevFeed: [Logit Laundering: Evading Data-Use Auditing in Large Language Models](<https://devfeed.tech/articles/logit-laundering-evading-data-use-auditing-in-large-language-models-122803.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09446>)

Author: Ruihan Hu, Wei Luo, Yu-Ming Shang, Jiakai Wang, Chuxuan Zhang, Mei Li, Xi Zhang, Meikang Qiu

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [data](<https://devfeed.tech/tags/data.md>), [data-ownership](<https://devfeed.tech/tags/data-ownership.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [detection](<https://devfeed.tech/tags/detection.md>), [estimator](<https://devfeed.tech/tags/estimator.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [evasion](<https://devfeed.tech/tags/evasion.md>), [framework](<https://devfeed.tech/tags/framework.md>), [inference](<https://devfeed.tech/tags/inference.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [next-token-prediction](<https://devfeed.tech/tags/next-token-prediction.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [utility](<https://devfeed.tech/tags/utility.md>), [watermark](<https://devfeed.tech/tags/watermark.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

Logit Laundering is an inference-time attack that weakens watermark-based audits for whether training data was used. It reshapes output logits using a plug-in statistical bias estimator, leaving model parameters unchanged, and scales suppression according to next-token confidence to balance audit evasion with model utility. The abstract reports that evaluations evaded existing state-of-the-art auditing methods while preserving downstream-task utility; it gives no specific methods, metrics, or results.

### Source excerpt

arXiv:2611.09446v1 Announce Type: new Abstract: Data-use auditing in Large Language Models (LLMs) aims to determine whether the given data has been used for training, thereby supporting data ownership safeguards. A dominant solution is data watermarking, which embeds traceable signals into training data and detects them in model outputs. Recent studies have proposed many watermarking-based auditing methods in LLMs. However, systematic investigation from an adversarial perspective remains limited: non-compliant LLM providers would use unauthorized copyrighted datasets for training their own models to enhance model capabilities, while attempting to evade detection by auditors, thereby avoiding legal liability. To fill this research gap, we propose a novel evasion framework, termed Logit Laundering, where the non-compliant LLM provider manipulates model outputs at inference time to undermine detectable watermark signals. Specifically, we design a logit reshaping module to suppress watermark-induced statistical signals by integrating a statistical bias estimator as a plug-and-play component. This design keeps the model parameters unchanged and performs manipulation only in the logit space, thereby enabling auditing evasion while minimizing the impact on model's learned knowledge. To further balance evasion effectiveness and model utility, we introduce a confidence-aware scaling module to adaptively calibrate the suppression strength based on next-token prediction confidence. Extensive evaluations show that Logit Laundering can successfully evade existing various state-of-the-art auditing methods, significantly attenuating detectable signals, while preserving model utility on downstream tasks. Our work highlights the need for exploring more robust auditing methods to guard against evasion behaviors.

## A Locally Tokenized Generative Model for Robust Time-Series Watermarking

DevFeed: [A Locally Tokenized Generative Model for Robust Time-Series Watermarking](<https://devfeed.tech/articles/a-locally-tokenized-generative-model-for-robust-time-series-watermarking-120076.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.19727>)

Author: Dongbin Kim, Geonwoo Shin, Yujin Choi, Soyeon Park, Jaewook Lee

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Time Series](<https://devfeed.tech/topics/time-series.md>), [Temporal data](<https://devfeed.tech/topics/temporal-data.md>)

Tags: [attack](<https://devfeed.tech/tags/attack.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [energy](<https://devfeed.tech/tags/energy.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [false-positive](<https://devfeed.tech/tags/false-positive.md>), [finance](<https://devfeed.tech/tags/finance.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [robust](<https://devfeed.tech/tags/robust.md>), [space](<https://devfeed.tech/tags/space.md>), [temporal](<https://devfeed.tech/tags/temporal.md>), [time-series](<https://devfeed.tech/tags/time-series.md>), [token](<https://devfeed.tech/tags/token.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>), [window](<https://devfeed.tech/tags/window.md>)

### AI overview

Existing time-series watermark detectors that use globally coupled re-encoding can shift non-watermarked samples' z-scores in either direction after post-editing, undermining thresholds calibrated on clean data. The paper proposes L-VQVAE, which produces each discrete token from a short contiguous time window, and LVQMark, which combines logit-bias watermark injection with robust re-encoding for detection after attacks. Across four finance, energy, and neuroimaging benchmarks, the authors report that the approach preserves generation quality while stabilizing detection power and false-positive behavior under post-editing.

### Source excerpt

arXiv:2608.19727v2 Announce Type: replace-cross Abstract: Watermarking is a central tool for provenance in generative models, yet its application to multivariate time series remains hindered by reliability failures under post-editing attacks. We show that existing detectors, which rely on globally coupled re-encoding, suffer from bidirectional drift of the null distribution: post-editing attacks can shift the $z$-score of non-watermarked samples in either direction, invalidating clean-calibrated thresholds. We argue that this instability is a property of the re-encoding, and that reliable detection requires each recovered unit to depend only on a bounded temporal neighborhood. Guided by this principle, we propose L-VQVAE, a generative model in which each discrete token is produced from a short contiguous window, and LVQMark, a watermarking method over this token space that combines logit-bias injection with robust re-encoding for attack-time detection. Experiments on four benchmarks spanning finance, energy, and neuroimaging show that our approach preserves generation quality while stabilizing both detection power and false-positive behavior under post-editing attacks.

## KeHoSu: Keyed Homoglyph Substitution for Authenticated Text Provenance over Copy-Paste Channels

DevFeed: [KeHoSu: Keyed Homoglyph Substitution for Authenticated Text Provenance over Copy-Paste Channels](<https://devfeed.tech/articles/kehosu-keyed-homoglyph-substitution-for-authenticated-text-provenance-over-copy-paste-channels-119923.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08046>)

Author: Andrei Alexandru Dr\\u{a}ghici, Sediola Ruko, Briland Hitaj, Dorjan Hitaj

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Cryptography](<https://devfeed.tech/topics/cryptography.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [authentication](<https://devfeed.tech/tags/authentication.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [channel](<https://devfeed.tech/tags/channel.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [document-metadata](<https://devfeed.tech/tags/document-metadata.md>), [fragments](<https://devfeed.tech/tags/fragments.md>), [invariant](<https://devfeed.tech/tags/invariant.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [platform](<https://devfeed.tech/tags/platform.md>), [platforms](<https://devfeed.tech/tags/platforms.md>), [provenance](<https://devfeed.tech/tags/provenance.md>), [unicode](<https://devfeed.tech/tags/unicode.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

KeHoSu encodes a platform, user and truncated authentication tag in Unicode homoglyph choices, with each carrier letter conveying one identifier bit. The keyed mark is designed for copy-paste channels that preserve code points but discard document metadata; its placement uses no positional state, making it invariant to sentence reordering up to a constant boundary layer and dependent on partial quotations only through their retained length. The paper characterizes capacity, removability and detectability and reports that the mark survived copy-paste across 18 common platforms. In evaluation, it attributed a majority of excerpts to the correct account at 178 letters for LLM-generated stories and 167 letters for standard English prose.

### Source excerpt

arXiv:2611.08046v1 Announce Type: new Abstract: Nowadays, large language models produce text at scale, and much of it is redistributed across platforms. A moderator who finds a suspicious comment, post or article may want to know which platform generated it and for which account. In this paper we study attribution over the channel through which such text actually travels, copy-paste, which discards document metadata but keeps Unicode code points. We present KeHoSu, which turns the confusable characters of homoglyph attacks into a keyed attribution channel: the variant chosen for each carrier letter encodes one bit of an identifier made of a platform field, a user field and a truncated authentication tag. Moreover, given that reshared text is often quoted in fragments, and the placement therefore keeps no positional state, we show that KeHoSu is invariant to sentence reordering up to a constant boundary layer and depends on a partial quotation only through the retained length. We characterize the code-point channel by its capacity, removability and detectability, and provide an achievable region that relates the three. We evaluate KeHoSu over a wide range of use cases and show that the mark survived copy-paste across 18 common platforms, attributing the majority of all excerpts to the correct account at 178 letters (about two sentences) on LLM-generated stories and 167 letters (about one sentence) on standard English prose.

## Majority Bit-Aware Watermarking For Large Language Models

DevFeed: [Majority Bit-Aware Watermarking For Large Language Models](<https://devfeed.tech/articles/majority-bit-aware-watermarking-for-large-language-models-119881.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2508.03829>)

Author: Jiahao Xu, Rui Hu, Zikai Zhang, Olivera Kotevska

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>)

Tags: [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [encoding](<https://devfeed.tech/tags/encoding.md>), [generation](<https://devfeed.tech/tags/generation.md>), [green](<https://devfeed.tech/tags/green.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llms](<https://devfeed.tech/tags/llms.md>), [quality](<https://devfeed.tech/tags/quality.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The paper proposes majority bit-aware encoding to reduce the trade-off between text quality and decoding accuracy in multi-bit LLM watermarking. Existing methods restrict the preferred-token "green list" to keep watermark signals detectable, which can degrade generated text. The proposed approach aims to preserve a strong signal with a larger green list, and is instantiated as MajorMark and MajorMark+, the latter optimized for long messages. The authors report experiments on state-of-the-art LLMs showing higher decoding accuracy and better text quality than prior baselines; the supplied abstract gives no quantitative results.

### Source excerpt

arXiv:2508.03829v4 Announce Type: replace-cross Abstract: The growing deployment of Large Language Models (LLMs) has raised concerns about their misuse in generating harmful or deceptive content. To address this issue, watermarking methods have been proposed to embed identifiable multi-bit messages into generated text for misuse tracing. However, existing methods often suffer from a fundamental trade-off between text quality and decoding accuracy. In particular, they have to restrict the size of the preferred token set (i.e., green list) during encoding to maintain a detectable watermark signal for decoding, which inevitably degrades generation quality. To improve this trade-off, we propose a novel message encoding paradigm called \textit{majority bit-aware encoding}, which relaxes the watermark signal strength from the green list size. This strategy allows for a strong watermark signal to be preserved in generated texts even when using a large green list. We introduce two instantiations of this paradigm: MajorMark and MajorMark$^{+}$, where the latter is specifically optimized for long messages. Extensive experiments on state-of-the-art LLMs demonstrate that our methods achieve higher decoding accuracy and superior text quality compared to prior baselines. The code is at https://github.com/JiiahaoXU/MajorMark.

## Value Over Language Model: Detecting Original Contribution in Writing

DevFeed: [Value Over Language Model: Detecting Original Contribution in Writing](<https://devfeed.tech/articles/value-over-language-model-detecting-original-contribution-in-writing-120080.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.00700>)

Author: Vibhhu Sharma, Thorsten Joachims, Sarah Dean

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [text-generation](<https://devfeed.tech/topics/text-generation.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [content](<https://devfeed.tech/tags/content.md>), [contribution](<https://devfeed.tech/tags/contribution.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [essays](<https://devfeed.tech/tags/essays.md>), [framework](<https://devfeed.tech/tags/framework.md>), [iclr](<https://devfeed.tech/tags/iclr.md>), [ideas](<https://devfeed.tech/tags/ideas.md>), [information](<https://devfeed.tech/tags/information.md>), [invariant](<https://devfeed.tech/tags/invariant.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-assisted](<https://devfeed.tech/tags/llm-assisted.md>), [llms](<https://devfeed.tech/tags/llms.md>), [news](<https://devfeed.tech/tags/news.md>), [original](<https://devfeed.tech/tags/original.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [reviews](<https://devfeed.tech/tags/reviews.md>), [surface](<https://devfeed.tech/tags/surface.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [text](<https://devfeed.tech/tags/text.md>), [writing](<https://devfeed.tech/tags/writing.md>)

### AI overview

VOLM estimates a document's contribution beyond what an LLM could produce from the task description alone, rather than judging surface wording. It extracts content at increasing levels of detail, asks an LLM to reconstruct the document from each representation, and compares those reconstructions with ones made from the task description alone. The method requires no training or labeled data. In evaluations across news articles, ICLR peer reviews, and argumentative essays, one VOLM implementation separated human writing from matched LLM-generated writing based on generic task descriptions and was substantially invariant to content-preserving transformations such as LLM reconstruction and round-trip translation. More constrained content extraction reduced residual differences between LLM-generated and humanized text, underscoring the challenge of separating informational contribution from style. The excerpt describes a framework and specific implementation, not a general performance guarantee.

### Source excerpt

arXiv:2609.00700v2 Announce Type: replace Abstract: LLMs have been rapidly adopted across writing tasks, prompting the development of tools for detecting LLM-generated text. Yet, these tools largely measure how much of a document's surface text was written by an LLM and aren't fundamentally designed to measure how much of the information content or ideas originated from the LLM itself rather than being supplied by the user in the prompt. In this work, we design a framework that measures how much value a person adds on top of what a language model could have easily produced by itself. The method requires no training or labeled data and never scores the document's surface text, insulating it from stylistic confounders. Instead, it extracts the document's content at increasing levels of granularity, uses an LLM to reconstruct the document from each partial representation, and compares these reconstructions with those produced from the task description alone. We call this framework Value Over Language Model (VOLM), which measures a document's contribution relative to a replacement-level document that an LLM could produce from the task description alone. We evaluate VOLM with a specific instantiation of this framework across three domains: news articles, ICLR peer reviews, and argumentative essays. VOLM separates human-authored documents from matched LLM-generated documents produced from generic task descriptions, while remaining substantially invariant to content-preserving transformations, including LLM-based reconstruction and round-trip translation. We further find that increasingly constrained content extractors reduce residual differences between LLM-generated and humanized text, demonstrating the importance of disentangling informational content from stylistic variation. We hope these results encourage further work on specialized instantiations of the framework and on assessing human contributions in LLM-assisted writing more generally.

## Reconciling Privacy with Provenance for Language Model Watermarks

DevFeed: [Reconciling Privacy with Provenance for Language Model Watermarks](<https://devfeed.tech/articles/reconciling-privacy-with-provenance-for-language-model-watermarks-118422.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06100>)

Author: Elise L. Han, Tatsunori Hashimoto

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [Security & Privacy](<https://devfeed.tech/topics/security-privacy.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [absolute](<https://devfeed.tech/tags/absolute.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [aggregate](<https://devfeed.tech/tags/aggregate.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [bias](<https://devfeed.tech/tags/bias.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [contribute](<https://devfeed.tech/tags/contribute.md>), [controversy](<https://devfeed.tech/tags/controversy.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [design](<https://devfeed.tech/tags/design.md>), [detection](<https://devfeed.tech/tags/detection.md>), [develop](<https://devfeed.tech/tags/develop.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [model](<https://devfeed.tech/tags/model.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [provenance](<https://devfeed.tech/tags/provenance.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

The paper proposes shifting language model watermarking from tracking individuals to estimating aggregate usage, aiming to protect privacy while preserving provenance estimates. Its Red-Green Sequence (RGS) watermark uses local differential privacy to limit how much evidence any individual text contributes to detection, while estimating population frequencies. The authors describe it as computationally efficient and report tests across models, sample sizes, and generation entropies. On C4 language modeling tasks, they report 0.0019 MSE and 0.0038 mean absolute bias at ε=1. For an essay task with 100 text samples, they report 0.0113 MSE and 0.004 mean absolute bias at ε=2. These results address aggregate frequency estimation; the abstract does not establish fine-grained individual tracking.

### Source excerpt

arXiv:2611.06100v1 Announce Type: cross Abstract: Watermarking has great potential for regulating the use of language models (LMs), but its deployment has been marred by controversy, as many users argue that fine-grained tracking of LM usage violates their privacy. This tension reduces the effectiveness of watermarks, as users engage in evasion strategies such as paraphrasing that can remove much of the watermark signal. We show that provenance and privacy can be reconciled if we shift the goal of watermark design from tracking individuals to estimating aggregate usage. We develop locally differentially private watermarks which provably limit the evidence that any individual text can contribute to detection, while still enabling accurate estimation of population frequencies. We introduce the Red-Green Sequence (RGS) watermark, a private and computationally efficient watermark that achieves accurate frequency estimation at high privacy levels across various models, sample sizes and generation entropies in settings that reflect realistic deployment scenarios. On C4 language modeling tasks, we attain $0.0019$ MSE with $0.0038$ mean absolute bias at a low $\varepsilon$ of 1, while in more challenging settings such as an essay writing task with only 100 text samples, we still find errors on the order of $0.0113$ MSE with $0.004$ mean absolute bias at $\varepsilon=2$.

## Query-efficient scan-and-search watermark localization

DevFeed: [Query-efficient scan-and-search watermark localization](<https://devfeed.tech/articles/query-efficient-scan-and-search-watermark-localization-114064.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02030>)

Author: Liyuan Wang, Xinsong Ma, Ziye Ma

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-generated-content](<https://devfeed.tech/tags/ai-generated-content.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [budget](<https://devfeed.tech/tags/budget.md>), [content](<https://devfeed.tech/tags/content.md>), [control](<https://devfeed.tech/tags/control.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [design](<https://devfeed.tech/tags/design.md>), [detection](<https://devfeed.tech/tags/detection.md>), [framework](<https://devfeed.tech/tags/framework.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llms](<https://devfeed.tech/tags/llms.md>), [localization](<https://devfeed.tech/tags/localization.md>), [query](<https://devfeed.tech/tags/query.md>), [scan](<https://devfeed.tech/tags/scan.md>), [search](<https://devfeed.tech/tags/search.md>), [watermark](<https://devfeed.tech/tags/watermark.md>)

### AI overview

BELL (Budgeted E-evidence Localization for LLM Watermarks) aims to reduce the cost of locating sparse watermarks in text. It first samples broadly across a document, then shifts its detection budget toward regions with stronger watermark evidence or a higher likelihood of containing watermarked content. It converts selected-region signals into e-values and applies a token-weighted e-BH procedure to localize regions. The abstract states that BELL controls token-weighted false discovery rate without assuming independence across regions, and reports better performance than competing methods across multiple LLMs, datasets, and watermarking schemes.

### Source excerpt

arXiv:2611.02030v1 Announce Type: new Abstract: Recently, the detection of AI-generated content has been of particular importance in various fields of our society. For accountability purposes, many large language models (LLMs) have implicitly watermarked their outputs. Localizing such watermarks requires a piece of text to undergo an additional verification process. Existing localization methods typically rely on dense token-level scoring, making verification costly even when an LLM-generated watermark only sparsely appears. To tackle this problem, we design a new adaptive allocation strategy that first maintains broad document coverage and then progressively concentrates the detection budget on regions that exhibit stronger watermark evidence or are more likely to contain watermarked content. We then transform the statistical signals from the selected regions into e-values and employ the proposed token-weighted e-BH procedure to determine the localized watermark regions. The resulting framework is called BELL (Budgeted E-evidence Localization for LLM Watermarks). Theoretically, BELL can control the token-weighted false discovery rate (FDR) without independence assumptions across regions. Finally, extensive experimental results demonstrate that BELL consistently outperforms competing methods across multiple LLMs, datasets, and watermarking schemes.

## Towards Fine-grained and Robust Detection of Human-AI Co-authorship

DevFeed: [Towards Fine-grained and Robust Detection of Human-AI Co-authorship](<https://devfeed.tech/articles/towards-fine-grained-and-robust-detection-of-human-ai-co-authorship-114545.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.01427>)

Author: Ziruo Zhao, Jie Fu, Wendy Hui Wang, Zining Zhu

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [watermarking](<https://devfeed.tech/topics/watermarking.md>), [data augmentation](<https://devfeed.tech/topics/data-augmentation.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Natural language processing](<https://devfeed.tech/topics/nlp.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [ai](<https://devfeed.tech/tags/ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data-augmentation](<https://devfeed.tech/tags/data-augmentation.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [detection](<https://devfeed.tech/tags/detection.md>), [discourse](<https://devfeed.tech/tags/discourse.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generation](<https://devfeed.tech/tags/generation.md>), [intellectual-property](<https://devfeed.tech/tags/intellectual-property.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llms](<https://devfeed.tech/tags/llms.md>), [meaning](<https://devfeed.tech/tags/meaning.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [plagiarism](<https://devfeed.tech/tags/plagiarism.md>), [provenance](<https://devfeed.tech/tags/provenance.md>), [robust](<https://devfeed.tech/tags/robust.md>)

### AI overview

The paper proposes a discrete-class framework for detecting degrees of human-AI co-authorship, distinguishing surface-level AI polishing from AI changes to discourse structure. It reports that adversarial paraphrasing substantially degrades detectors trained on benign data by blurring their decision boundaries. Training with meaning-preserving adversarial paraphrases improves robustness across multiple datasets and shows good transfer across generation models.

### Source excerpt

arXiv:2611.01427v1 Announce Type: cross Abstract: The widespread adoption of large language models (LLMs) is reshaping how textual content is produced, complicating text provenance, and raising concerns for misinformation, plagiarism, and intellectual property. While prior work on AI-generated text detection has progressed, one-dimensional human-AI co-authorship scores lack signals directly relevant to ``AI involvement''. It remains unclear how adversarial paraphrasing distorts detection signals. In this work, we formulate a discrete-class text provenance detection framework. This framework allows an abstraction of realistic co-authorship scenarios into distinct degrees of AI involvement, differentiating the AI polishing on the surface level from the AI updating the discourse structures. This framework reveals the mechanisms of how adversarial paraphrasing blurs the decision boundaries for co-authorship detectors. To improve robustness of the detectors, we augment training data with adversarial paraphrases generated by the state-of-the-art paraphrasing attack model, exposing detectors to challenging meaning-preserving rewrites. Extensive experiments across multiple datasets and generation models demonstrate that adversarial paraphrasing substantially degrades detectors trained only on benign data, while adversarial data augmentation leads to consistent robustness gains and good cross-generator transferability. Overall, our findings highlight the value of discrete-class formulation and principled data augmentation for building reliable detectors of human-AI co-authorship.

## The latest AI news we announced in October 2026

DevFeed: [The latest AI news we announced in October 2026](<https://devfeed.tech/articles/the-latest-ai-news-we-announced-in-october-2026-112629.md>)

Original publisher: [Read original article](<https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-october-2026/>)

Author: Blog Team

Published: 2026-11-02T17:45:00Z

Content type: news

Language: en

Sources: [Gemini Models](<https://devfeed.tech/sources/gemini-models.md>), [Google DeepMind](<https://devfeed.tech/sources/google-deepmind.md>), [Google Research](<https://devfeed.tech/sources/google-research-2.md>), [Quantum computing](<https://devfeed.tech/sources/quantum-computing.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [generative-ai-tools](<https://devfeed.tech/topics/generative-ai-tools.md>), [watermarking](<https://devfeed.tech/topics/watermarking.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-news](<https://devfeed.tech/tags/ai-news.md>), [fitbit](<https://devfeed.tech/tags/fitbit.md>), [gemini-app](<https://devfeed.tech/tags/gemini-app.md>), [gemini-models](<https://devfeed.tech/tags/gemini-models.md>), [google-deepmind](<https://devfeed.tech/tags/google-deepmind.md>), [google-research](<https://devfeed.tech/tags/google-research.md>), [latest](<https://devfeed.tech/tags/latest.md>), [news](<https://devfeed.tech/tags/news.md>), [none](<https://devfeed.tech/tags/none.md>), [october](<https://devfeed.tech/tags/october.md>), [october-2026](<https://devfeed.tech/tags/october-2026.md>), [posts](<https://devfeed.tech/tags/posts.md>), [quantum-computing](<https://devfeed.tech/tags/quantum-computing.md>), [the-latest](<https://devfeed.tech/tags/the-latest.md>)

### AI overview

Google's October 2026 announcements span AI models, consumer tools, research and infrastructure. Gemini 4 Argon rolled out globally for complex workflows; Gemini 4 Carbon, described as its strongest model yet, is in limited pre-release testing with U.S. government and trusted cyber defenders while safety guardrails are refined. Other launches include Genie 4's real-time interactive simulations with natural-language control and native audio, and a global English expansion of SynthID Detector for checking AI-generated images, video and audio, including content from named industry partners. Android's new opt-in Messages protection uses Private AI Compute and cloud Gemini models to detect sophisticated scams. Google also reports AMIE helped clinicians prepare in 75% of evaluated cases and matched doctors' final differential diagnoses 90% of the time. In quantum computing, a Willow processor study found magic state cultivation cut errors roughly 40-fold. The roundup also describes an on-device multimodal embedding model, experimental creative and learning tools, and a satellite test of AI workloads in orbit; these remain at varying stages of availability or testing.

### Source excerpt

Animated carousel of images from AI posts in October

[Next page](<https://devfeed.tech/topics/watermarking.md?cursor=WyIyMDI2LTExLTAyVDE3OjQ1OjAwKzAwOjAwIiwgIjgzMDAxZjJmLWNlZWYtNDIyMi1iZWJiLTZlOGNlZmQ1OTIyNiJd>)