Anthropic Hands Petri to Meridian Labs, Betting Open Governance Keeps Alignment Tooling Credible

5. Anthropic Hands Petri to Meridian Labs, Betting Open Governance Keeps Alignment Tooling Credible

Anthropic has donated Petri, its open-source alignment evaluation tool, to Meridian Labs for independent stewardship. The transfer was announced May 6, 2026, alongside a major update to Petri developed jointly with Meridian Labs. The update improves the adaptability, realism, and depth of Petri's test suite. No acquisition price was involved; this is a governance transfer, not a sale.

The move exposes a structural tension every frontier lab faces: alignment tooling built and maintained in-house carries an implicit conflict of interest. Anthropic's own Claude models are the most obvious downstream subject of any evaluation Petri runs. By moving Petri to Meridian Labs, Anthropic separates the tool's development roadmap from its own commercial incentives. That separation matters competitively. OpenAI's Evals framework and Google DeepMind's internal safety benchmarks remain lab-controlled, which makes third-party researchers and regulators skeptical of the results those tools produce. An independently governed Petri can attract contributions and credibility that a lab-owned repo cannot. Meridian Labs gains a high-profile asset and a direct line into the alignment research community; Anthropic gains the ability to cite Petri results without the obvious objection that it graded its own homework.

Watch whether other frontier labs follow with similar transfers, or whether they treat independent governance as a reputational threat rather than an asset. The EU AI Act's conformity assessment requirements and the UK AI Safety Institute's evaluation mandates both create pressure toward third-party tooling. Petri under Meridian Labs is now positioned to fill that role. If it does, the donation becomes a quiet land-grab for evaluation authority, not just an act of open-source generosity.

Source: @AnthropicAI on X