TechFlow reported on September 21, citing The Information, that a source familiar with the matter revealed that long before OpenAI’s technology triggered a series of cybersecurity incidents and industry insiders issued stern warnings, the company was already negotiating a legally binding agreement with Anthropic, under which both parties would conduct mutual stress tests on their AI models.
It remains unclear whether they have finalized an agreement. This concept of reciprocal testing aligns with a proposal made last week by SpaceX CEO Elon Musk at the All-In Summit. At the event, Musk suggested that competing AI laboratories should evaluate each other's models for security vulnerabilities via a peer-review-style process prior to commercial release.





