Alexander Panfilov

Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLMs

Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols

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

ASIDE: Architectural Separation of Instructions and Data in Language Models

A Realistic Threat Model for Large Language Model Jailbreaks

An Interpretable N-gram Perplexity Threat Model for Large Language Model Jailbreaks