Summary

Research engineer/scientist role in alignment science at anthropic, focused on exploratory experimental research for ai safety. You will build and run machine learning experiments, including reinforcement learning and evaluation tooling for llm jailbreaks, with an emphasis on risks from powerful future systems. The london team follows a hybrid policy requiring at least 25% time in london and occasional travel to san francisco, and interviews are conducted in python.

Responsibilities

• Build and run machine learning experiments
• Test safety techniques against subversion attempts
• Run multi-agent reinforcement learning experiments
• Build tooling to evaluate llm jailbreaks
• Write scripts and prompts for safety evaluations
• Contribute ideas and figures to research outputs
• Run experiments feeding into responsible scaling policy