Research Engineer Safety and Alignment
CognitionVisit Cognition website
Cognition is an applied AI company that operates Devin, an autonomous software-engineering agent.
Maintainer signals as of 9/25/2026
San Francisco, United States
Funding history
About Cognition
Cognition builds AI agents and models for software engineering. Its flagship product, Devin, can plan, write, test, and ship code in a customer’s existing codebase and tools.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will build and run safety evaluations and red-team an autonomous AI agent. You will develop alignment methods, work on permissions and oversight in the agent harness, run experiments, and contribute to external safety research through papers, evaluations, and open methods.
Requirements
- Safety, alignment, evaluation, red-teaming, or interpretability research experience.
- Knowledge of agentic failure modes.
- Proficiency in Python and PyTorch or JAX.
- Ability to build evaluation infrastructure and run experiments at scale.
- Experience designing rigorous experiments and reporting results.
- Experience with adversarial testing.
- Experience at a frontier AI lab, applied AI company, or developer tools company.
- A PhD in a relevant discipline or equivalent industry research experience.
Responsibilities
- Design and run agentic safety evaluations.
- Integrate evaluations into every model and product release.
- Red-team the agent and its harness.
- Turn discovered failures into fixes and regression tests.
- Develop alignment methods with post-training teams.
- Shape permissions, oversight, and escalation in the agent harness.
- Contribute to papers, evaluations, and open safety methods.