LeakGauge detects LLM context-leakage via behavior gauges
chaumian · x · 2026-08-19
A paper on arXiv introduces LeakGauge, a method to measure signals of context-leakage attacks in LLMs. Adversarial inputs can trick models into disclosing external contexts like system prompts. LeakGauge appends a suffix to gauge leakage behavior and maps prefill token probabilities to a risk score. The study finds that a content-agnostic gauge yields more robust signals than direct content probing. Tested on 11 LLMs including GLM-5.2 and Kimi-K3, it achieved an AUROC range of 0.944--0.996 on unseen attacks.
More from Safety
- Blogger Confused by Public Criticism of Anthropic's Watermarking — repligate · 2026-08-19
- Claude automatically downgrades queries with specific words, raising safety concerns — 1a3orn · 2026-08-19
- Reddit thread: ignoring safety will win the RSI race because humans in the loop are slow — TwoFluid4446 · 2026-08-19
- Blog: AI Accelerates Arms Race Between Fraudsters and Honest Researchers — sebkrier · 2026-08-19
- Tencent Evaluates DeepSeek Harness Resistance to Indirect Prompt Injection — tencent · 2026-08-19
- PANDA: Scalable ZKPs for Private Neural Network Guarantees — chaumian · 2026-08-19