OpenAI and Anthropic discussed testing each other's models, The Information says
Five financial wire accounts relayed the report on Monday. All of them rest on the same outlet, and neither company has commented.
OpenAI and Anthropic came close earlier this year to an agreement under which each would test the other's models for vulnerabilities and hidden safety risks, The Information reported. Five accounts relayed the report on X on Monday, and no other outlet has matched it.
The relays differ on what the arrangement would have been. The account wallstengine described a legally binding deal that would let each company test the other's models. DeItaone said it is unclear whether the deal was finalised before OpenAI encountered incidents involving unreleased models.
None of the five accounts reported the story itself. They are financial wire feeds that summarise other outlets, and each credits The Information. The paper has not read the report, which sits behind a subscription, and is working from the summaries.
The Information said separately that AI companies are promising to bring outside safety evaluators deeper into their operations as scrutiny of AI risks grows. It said critics are questioning whether some of the groups positioned to serve as watchdogs are independent enough.
Neither OpenAI nor Anthropic has commented on the report. Whether the two companies signed anything is unclear, and the only relay to address that question said it could not say. Until one of them speaks, the story rests on a single outlet.