AI/TLDR

Anthropic · 2026-05-07 · major

Anthropic Donates Petri to Meridian Labs — Petri 3.0 Splits Auditor From Target, Adds Bloom Integration

Anthropic transfers its open-source alignment-auditing toolbox Petri to nonprofit Meridian Labs and ships Petri 3.0, which decouples the auditor from the target model and plugs into Bloom for deeper behaviour assessments.

Anthropic illustration accompanying the Petri donation announcement
Anthropic

Anthropic hands Petri to an independent nonprofit and ships a 3.0 release that lets you swap auditor and target models without rewriting the harness.

Key specs

LicenseMIT
GitHub stars1,122
Default scoring dimensions36

What is it?

Petri is an open-source toolbox of automated alignment audits — agents that prod a target model across multi-turn scenarios looking for deception, sycophancy, and other concerning behaviours. Anthropic shipped it in October 2025; today they're donating it to Meridian Labs, an independent nonprofit, so the tool can be governed outside any single AI lab. Alongside the donation they're shipping Petri 3.0.

How does it work?

Petri 3.0 splits the auditor and target into independent components that talk through a defined interface, instead of the auditor manipulating the target's message history directly. Three add-ons land with the split: Dish runs audits using the model's real production system prompt and scaffold (defeating eval-awareness), Bloom plugs in Anthropic's complementary alignment tool for deeper behaviour-specific probes, and a rollback feature replays events to restore identical target states for repeatable runs.

Why does it matter?

Petri results are already cited in UK AI Security Institute evaluations and in Anthropic's own model cards. Independent governance is what regulators and labs have been asking for — if Petri is going to play a role in pre-release safety review, it cannot live inside the lab whose models it audits. The 3.0 architecture also makes it easier for outside researchers to audit closed-API models without lab cooperation.

Who is it for?

alignment researchers, AI safety teams, regulators

Try it

pip install inspect-petri

Sources · 3 outlets

Tags

  • alignment
  • evaluation
  • open-source
  • anthropic
  • meridian-labs
  • governance
  • petri
  • ai-safety
  • auditing
  • inspect

← All releases · Learn AI