How AI describes Petri

Petri is a tool for rapidly testing alignment hypotheses end-to-end by generating realistic audit scenarios and orchestrating multi-turn audits using an auditor model, a target model, and a judge model. It simulates tools and rollbacks to probe behaviors and then scores transcripts with a judge model under a consistent rubric. It works with any Inspect-compatible models, provides a quick-start workflow, and emphasizes responsible use when probing for potentially harmful content.