Skip to content
AI HOT (Curated Pool)

Donating the Open-Source Alignment Tool Petri

捐赠开源对齐工具 Petri

Anthropic transferred the open-source alignment testing tool Petri to Meridian Labs to preserve independence and credibility. Petri 3.0 separates auditor and target models, adds Dish for real prompts and deployment settings, and integrates Bloom.

Why it matters: HKR-H/K/R all pass: the independent donation is a real hook, Petri 3.0 and Dish add testable mechanisms, and audit credibility resonates. Anthropic open-source safety tooling is strong, but below a model-release-level event.

Read the original ↗Export Markdown