4 papers
PatchBench: Measuring Collateral Damage in Activation Patching
Alexi Canesse, Mathis Le Bail, Maël Jenny +3
An LLM safety patch can pass a benchmark while still being a poor repair. This risk is especially acute for jailbreak repairs, where the goal is to correct a specific unsafe behavi…
RAISED: Self-Distillation for Robustness to Prompt Injection in LLM Agents
Mohamed Dhouib, Clement Elliker, Alexi Canesse +6
Tool-using language-model agents are vulnerable to indirect prompt injection because they must act on untrusted external content. Existing training-time defenses can reduce attack…
Reformulation-Contrastive Learning for Mixed Integer Programs
Ousema Bouaneni, Mathis Le Bail, Clément Elliker +2
Mixed-integer linear programs (MILP) model many real-world decision problems, motivating machine-learning methods that exploit recurring structure to accelerate MILP solving. MILPs…
Activation Surgery: Jailbreaking White-box LLMs without Touching the Prompt
Maël Jenny, Jérémie Dentan, Sonia Vanier +1
Most jailbreak techniques for Large Language Models (LLMs) primarily rely on prompt modifications, including paraphrasing, obfuscation, or conversational strategies. Meanwhile, abl…