OpenAI, Anthropic Discussed Binding Cross-Stress-Testing Pact for AI Models

A binding cross-stress-testing pact between rival labs would be a notable shift from voluntary, self-reported safety evaluations toward third-party verification, giving each company visibility into the other's models that neither currently has. Because no agreement was reached and the discussions predate the recent cybersecurity incidents and industry warnings, the open question is whether those events revive the talks or push labs toward unilateral safety measures instead. For the broader AI and tokenized-asset space, the episode signals that safety scrutiny is increasingly being framed around enforceable commitments rather than stated principles.