# Escalate or Close: How Fast-Tier Models Handle Held-Out Web Attack Families in Two-Tier Triage

DevFeed: [Escalate or Close: How Fast-Tier Models Handle Held-Out Web Attack Families in Two-Tier Triage](<https://devfeed.tech/articles/escalate-or-close-how-fast-tier-models-handle-held-out-web-attack-families-in-two-tier-triage-119928.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08229>)

Author: Maciej Kisielewicz, Micha{\\l} K\\k{e}dziora, Ngoc Thanh Nguyen

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: [alert triage](<https://devfeed.tech/topics/alert-triage.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>), [AI Models](<https://devfeed.tech/topics/ai-models.md>)

Tags: [attack](<https://devfeed.tech/tags/attack.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [fast](<https://devfeed.tech/tags/fast.md>), [models](<https://devfeed.tech/tags/models.md>), [owasp](<https://devfeed.tech/tags/owasp.md>), [path-traversal](<https://devfeed.tech/tags/path-traversal.md>), [triage](<https://devfeed.tech/tags/triage.md>), [web](<https://devfeed.tech/tags/web.md>)

## AI overview

In two-tier web-request triage, attacks closed by the fast tier never reach the slower reviewer. With each of seven attack families held out of training and thresholds, supervised detectors could mistake many unseen attacks for benign: ModernBERT assigned 37-60% of held-out path-traversal requests probabilities below 0.05 and escalated 50-81% within a 5% benign budget, versus nearly all when traversal was known. Hosted decision models Jev and Clef escalated at least 97.6% of every held-out family at that budget, though their undisclosed training data may include the public attacks. Shape controls matter: Jev also escalated 37% of benign exploit-request twins, while detectors trained on older exploits learned scanner-request patterns. A search using Jev's probabilities got 43% of held-out attacks past it within 300 queries; results varied on repeat, with those requests escalated in 60% of Jev's answers.

## Source excerpt

arXiv:2611.08229v1 Announce Type: new Abstract: In two-tier triage a fast model decides most web requests and escalates the rest to a slow tier, so an attack the fast tier closes is never examined. We hold each of seven attack families out of training, recalibration and thresholds, splice the payloads into parameters of real browsing traffic, and compare zero-shot decision models, which return a probability for a typed question, with supervised detectors, a local language model, the OWASP Core Rule Set and an anomaly baseline. Detectors trained on the known families can give much of a held-out family a confident benign verdict: over 10 training runs, ModernBERT gives 37% to 60% of held-out path traversal a probability below 0.05 and escalates only 50% to 81% of it within a 5% benign budget, against nearly all of the same requests when traversal is known. Jev and Clef, two hosted decision models whose budget decisions do not depend on the held-out family, escalate at least 97.6% of every family within the same budget, but their undisclosed training data may contain these public families. Tests on recent exploits need controls for request shape: Jev escalates 37% of benign twins (exploit requests with the attack replaced by legitimate values), and detectors trained on older exploits learn what scanner requests look like. A search guided by Jev's probability gets 43% of held-out attacks past Jev within 300 queries, but Jev's answers vary: asked again, it escalates those requests in 60% of its answers.