Red-teaming the defences of frontier AI.

Recent work: Stealing Reasoning Traces from Proprietary LLM APIs