Research Engineer / Scientist, Safeguards (San Francisco)

About the role

The Safeguards Research Team is part of the Alignment Science team , and conducts critical safety research and engineering to ensure AI systems can be deployed safely. As part of Anthropic's broader safeguards organization, we work on both immediate safety challenges and longer-term research initiatives, with projects spanning jailbreak robustness, automated red-teaming, monitoring techniques, and applied threat modeling. We prioritize techniques that will enable the safe deployment of more advanced AI systems (ASL-3 and beyond), taking a pragmatic approach to fundamental AI safety challenges while maintaining strong research rigor.

You take a pragmatic approach to running machine learning experiments to help us understand and steer the behavior of powerful AI systems. You care about making AI helpful, honest, and harmless, and are interested in the ways that this could be challenging in the context of human-level capabilities. You could describe yourself as both a scientist and an engineer. Youll both focus on risks from powerful future systems (like those we would designate as ASL-3 or ASL-4 under our Responsible Scaling Policy), as well as better understanding risks occurring today. You will work in collaboration with other teams including Interpretability, Fine-Tuning, and the Frontier Red Team.

These papers give a simple overview of the topics the team works on: Best-of-N Jailbreaking , Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats , Rapid Response: Mitigating LLM Jailbreaks with a Few Examples , Many-shot Jailbreaking , When Do Universal Image Jailbreaks Transfer Between Vision-Language Models?

Note: Currently, the team has a preference for candidates who are able to be based in the Bay Area. However, we remain open to any candidate who can travel 25% to the Bay Area.

Representative projects:

• Testing the robustness of our safety techniques by training language models to subvert our safety techniques, and seeing how effective they are at subverting our interventions.
• Run multi-agent reinforcement learning experiments to test out techniques like AI Debate.
• Build tooling to efficiently evaluate the effectiveness of novel LLM-generated jailbreaks.
• Write scripts and prompts to efficiently produce evaluation questions to test models reasoning abilities in safety-relevant contexts.
• Contribute ideas, figures, and writing to research papers, blog posts, and talks.
• Run experiments that feed into key AI safety efforts at Anthropic, like the design and implementation of our Responsible Scaling Policy.

You may be a good fit if you:

• Have significant software, ML, or research engineering experience
• Have some experience contributing to empirical AI research projects
• Have some familiarity with technical AI safety research
• Prefer fast-moving collaborative projects to extensive solo efforts
• Pick up slack, even if it goes outside your job description
• Care about the impacts of AI

Strong candidates may also:

• Have experience authoring research papers in machine learning, NLP, or AI safety
• Have experience with LLMs
• Have experience with reinforcement learning
• Have experience with Kubernetes clusters and complex shared codebases