# ADTGuard: Adaptive Depth-Time Safeguard for Streaming Inference in Large Language Models

DevFeed: [ADTGuard: Adaptive Depth-Time Safeguard for Streaming Inference in Large Language Models](<https://devfeed.tech/articles/adtguard-adaptive-depth-time-safeguard-for-streaming-inference-in-large-language-models-103155.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.36574>)

Author: Chuming Wei, Churui Yang, Zihan Zhang, Ruien Li, Wusheng Zhang, Yan Zhang, Guangwen Yang

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

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

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [detection](<https://devfeed.tech/tags/detection.md>), [inference](<https://devfeed.tech/tags/inference.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [latency](<https://devfeed.tech/tags/latency.md>), [layer](<https://devfeed.tech/tags/layer.md>), [risk](<https://devfeed.tech/tags/risk.md>), [safety](<https://devfeed.tech/tags/safety.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [supervision](<https://devfeed.tech/tags/supervision.md>), [token](<https://devfeed.tech/tags/token.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [training](<https://devfeed.tech/tags/training.md>)

## AI overview

ADTGuard checks each generated token before release and models safety risk across both layers and time. Its Depth-Time Safety Dynamics (DTSD) combines safety evidence across layers at each token to guide risk accumulation over the sequence, while the Causal Depth Router (CDR) adapts which layers it observes to the current prefix. Because training labels mark responses rather than harmful-token onset, Risk-Onset Target Refinement (RTR) progressively constructs token-level targets for both components. The abstract reports strong results across multiple safety benchmarks, matching or surpassing state-of-the-art safeguards with 16M trainable parameters and competitive per-token latency; it does not provide benchmark-specific results or latency figures.

## Source excerpt

arXiv:2610.36574v1 Announce Type: new Abstract: Large Language Models (LLMs) can generate fluent responses to diverse user requests, yet streaming may expose users to harmful content before generation is complete. Existing guardrails face two limitations in this setting. First, post-hoc moderation at response or chunk boundaries delays intervention, risking harmful exposure when content is released before verification. Second, even with token-level detection, coarse response-level supervision does not specify when harmful content begins, while fixed-layer readouts may miss relevant safety signals at other depths. To address these limitations, we introduce ADTGuard, which checks each token before release, progressively refines token-level training targets, and adapts layer selection to the current prefix. ADTGuard models risk across depth and time through Depth-Time Safety Dynamics (DTSD), which integrates safety evidence across layers at each token and uses it to guide risk accumulation over successive tokens. To adapt the choice of observation layers to the current prefix, we propose the Causal Depth Router (CDR). With only response-level labels available, we further develop Risk-Onset Target Refinement (RTR) to progressively construct token-level training targets for both components. Through extensive experiments across multiple safety benchmarks, we demonstrate that ADTGuard consistently achieves strong streaming detection performance, matching or surpassing state-of-the-art safeguards while requiring only 16M trainable parameters and maintaining competitive per-token latency.