Definition
Petri is Anthropic’s open-source tool for auditing model behavior in simulated multi-turn environments, used to elicit and measure agentic misalignment case studies.
Key Points
- Auditor model plays environment/operators; target model is evaluated
- Used for Summer 2026 agentic misalignment portfolio frequency sweeps (~20 runs/model/scenario)
- Not a substitute for application security review or production monitoring