We’re donating Petri, our open-source alignment tool, to @meridianlabs_ai, so it

Anthropic has announced the donation of Petri, its open-source AI alignment testing tool, to Meridian Labs AI, signaling a strategic decision to transfer stewardship of the project to an independent organization better positioned to sustain its long-term development. Alongside the handoff, Anthropic and Meridian Labs jointly released a significant update to the tool, improving three core dimensions of its testing capabilities: adaptability, realism, and depth. The move represents a deliberate effort to ensure Petri's continued evolution outside of Anthropic's direct control, potentially allowing it to serve a broader community of AI safety researchers and developers without being constrained by the priorities of a single commercial entity. Petri functions as an alignment evaluation framework, designed to probe AI systems for behavioral properties relevant to safety and alignment. The enhancements to adaptability suggest the tool can now be more readily configured to test diverse model architectures or deployment contexts, while improvements to realism likely mean that test scenarios more closely mirror actual deployment conditions rather than synthetic edge cases. Greater depth in testing implies more thorough and granular interrogation of model behavior, which is critical for detecting subtle misalignment that surface-level evaluations might miss. These improvements collectively make Petri a more credible instrument in the broader alignment research toolkit. The decision to open-source and then donate Petri reflects a pattern increasingly visible across the AI safety ecosystem, wherein foundational tools and methodologies are deliberately decoupled from proprietary interests to encourage wider adoption and independent scrutiny. Anthropic has historically emphasized that AI safety is not a competitive advantage to be hoarded, and this transfer operationalizes that principle. By handing the project to Meridian Labs, Anthropic reduces the risk that Petri's development stalls if internal resource priorities shift, while also signaling confidence in the broader research community's capacity to carry the work forward responsibly. The broader significance of this move lies in the maturation of AI alignment as a technical discipline. The existence of purpose-built, openly available tools like Petri indicates that alignment evaluation is increasingly treated as an engineering discipline with reproducible methods, not merely a philosophical or theoretical exercise. As frontier AI systems grow more capable, the demand for rigorous, standardized evaluation frameworks grows in parallel. Transferring Petri to an independent custodian like Meridian Labs positions the tool to become a shared infrastructure resource for the alignment research community, potentially enabling cross-organizational benchmarking and collaborative refinement in ways that would be structurally difficult if the tool remained under Anthropic's exclusive ownership.