Maksym Andriushchenko

Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols

Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLMs

Capability-Based Scaling Trends for LLM-Based Red-Teaming

Monitoring Decomposition Attacks with Lightweight Sequential Monitors

AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents

Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks

Does Refusal Training in LLMs Generalize to the Past Tense?

Critical Influence of Overparameterization on Sharpness-aware Minimization

Is In-Context Learning Sufficient for Instruction Following in LLMs?