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

Summary

Two-tier web triage, where a fast model closes most requests and escalates the rest, is tested on seven held-out attack families. Supervised detectors such as ModernBERT give many held-out path traversal requests a confident benign verdict, escalating only 50% to 81% of them within a 5% benign budget. Two hosted zero-shot decision models, Jev and Clef, escalate at least 97.6% of every family within the same budget, though their undisclosed training data may include these public families.

Related items

• MediumCPython flaw exposes AI inference servers to sandbox escape attacksSimilar attack · CSO Online
• MediumCVE-2026-88979: Kubeflow Pipelines TensorBoard proxy forgeable token via default HMAC secretSimilar attack · NVD/CVE Database
• InfoEvaluating Parameter-Efficient Fine-Tuning Attacks Against Non-Transferable Learning DefensesSimilar attack · Elsevier Security Journals
• MediumResearchers Bypass ChatGPT Lockdown Mode to Exfiltrate Chats Without User InteractionSimilar attack · The Hacker News
• MediumChatGPT Lockdown Mode Bypass: Phishing, Malware Delivery, and Silent Data ExfiltrationSimilar attack · Check Point Research