αzimuth λabs
AI SECURITY RESEARCH

Mapping Frontier
AI Risks

Azimuth Labs is an independent research nonprofit working to keep AI models from enabling biological harm.

Supported by Anthropic's AI for Science program · Center for AI Safety (CAIS)

Research Focus

Defense-in-depth research to keep AI models from enabling biological harm.

Pre-training Data Filtering

Researching how filtering pre-training data can keep hazardous biological knowledge out of models from the start.

Tamper-Resistance for Open-Weight Models

Developing safeguards for open-weight models that hold up after release, including against attempts to remove them.

Multi-Agent Safety Evaluation

Evaluating whether systems of interacting AI agents could enable biological harm, and testing safeguards against it.