PETRI

Summary from the TLDR AI newsletter

Petri is an alignment auditing agent for realistic hypothesis testing. It allows researchers to test new hypotheses in minutes instead of building bespoke evals over weeks. Petri autonomously crafts environments, runs multi-turn audits, and scores transcripts to surface concerning behavior.