Partly True
OpenAI and Anthropic are close to reaching an agreement under which the two rival AI developers would cross-validate each other's AI models, though the deal has not yet been finalized and undisclosed models would fall outside its scope, according to information reviewed in connection with the claim.
The proposed arrangement remains at the near-agreement stage, and whether it will ultimately be concluded is still unclear. The claim that a deal is "imminent" therefore cannot be established as fully factual, but neither does the available evidence contradict it.
Under the framework as described, models that have not been publicly released would be excluded from the cross-validation process. Anthropic's alignment research materials reviewed in connection with the matter include a chart labeled "Cooperation with human misuse scores plot," indicating the companies have examined safety-related evaluation measures in a collaborative context.
The talks center on mutual evaluation of model safety, with Anthropic's published alignment work suggesting the two firms have shared findings on risks such as cooperation with human misuse. Details of the scope, timeline and enforcement of any agreement remain unconfirmed.
Verdict: Partly True