Research Engineer, Safety & Alignment

The role

You will be a founding member of the safety team building evaluations and red-teaming for Devin. You will develop alignment methods and work with the agent team on how the agent plans, acts, and asks for help.

What you will do

• Design and run evaluations for failure modes like unsafe actions, prompt injection, data exfiltration, sandbox escape, reward hacking, and misuse.
• Attack the agent and its harness systematically through red-teaming.
• Develop alignment methods including reward modeling, preference data, and constitutional approaches.
• Partner with the agent team on permissions, oversight, and escalation.
• Contribute to external safety work through papers, evals, and open methods.

Required qualifications

• Safety or alignment research experience in evaluations, red-teaming, alignment, or interpretability.
• Agentic systems understanding regarding tool misuse, specification gaming, long-horizon drift, and adversarial inputs.
• Proficiency in Python and PyTorch or JAX.
• Ability to build eval infrastructure and run experiments at scale.
• Empirical rigor in designing experiments and reporting results.
• Comfort with adversarial thinking.
• PhD in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline; or equivalent industry research experience.

Nice to have

• Prior experience at a frontier AI lab, applied AI company, or developer tools company.

About the company

Cognition is an applied AI lab building end-to-end software agents. They are the makers of Devin, the first AI software engineer.