OpenAI and Anthropic, two of the world's largest artificial intelligence labs, negotiated a legal agreement to test each other's models for safety risks, according to reporting from The Information. The two companies held a joint safety evaluation in August 2025, during which OpenAI tested Anthropic's Claude Opus 4 and Claude Sonnet 4, while Anthropic tested OpenAI's GPT-4.0, GPT-4.1, and other models. It remains unclear whether the two companies finalized the agreement, though both have publicly called for industry-wide efforts to pace AI development.

Why Would Competitors Test Each Other's AI Models?

The proposed agreement reflects a broader shift in how AI companies approach safety. Rather than evaluating their own systems in isolation, OpenAI and Anthropic explored a mutual verification approach. This kind of cross-company testing could help identify risks that internal teams might miss, and it signals a willingness to share sensitive technical information in the name of safety. The move comes as AI safety concerns have become increasingly prominent in industry conversations and regulatory discussions.

The timing of these talks is significant. In September 2026, Anthropic CEO Dario Amodei called for the industry to slow down AI development, citing potential safety risks to humanity. OpenAI CEO Sam Altman, SpaceX CEO Elon Musk, and DeepMind chair Dennis Hassabis all backed the idea of such a slowdown. This shared concern about safety appears to have motivated the cross-testing proposal, even though the two companies are fierce competitors in the commercial AI market.

How Are AI Companies Approaching Safety Testing?

• Internal Benchmarking: Both Anthropic and OpenAI use proprietary benchmarks to evaluate their models before release, measuring performance on tasks like coding, reasoning, and real-world workflows.
• Third-Party Partnerships: Anthropic partnered with Accenture on independent evaluation of frontier AI models, with the two companies expecting to invest over $1 billion in building testing capacity over the next five years.
• Cross-Company Verification: The proposed OpenAI-Anthropic agreement would allow each company to audit the other's models, creating a mutual accountability mechanism that neither company could achieve alone.

Anthropic has been particularly vocal about the need for independent oversight.

"To be clear, independent embedded evaluators do not reduce our accountability, but help to make it more verifiable. The safety of our models remains our responsibility," Anthropic stated in its press release about its partnership with Accenture.

Anthropic, Company Statement

What Does This Mean for the Broader AI Industry?

The proposed agreement between OpenAI and Anthropic suggests that even as companies compete fiercely on model performance and pricing, they may be willing to collaborate on safety. This mirrors calls from other industry leaders for coordinated action. Elon Musk, founder of xAI, has suggested that American AI creators and Chinese competitors should test each other's models, according to remarks he made at the All-In Summit in Los Angeles in September.

However, not all tech executives support this collaborative approach. Nvidia CEO Jensen Huang has argued that new regulations or antitrust exemptions are unnecessary, stating that "every lab has the responsibility and incentive to move at the pace required to train its models safely." Meta CEO Mark Zuckerberg echoed this sentiment, suggesting that individual companies should take their own actions to ensure safe development.

The fact that OpenAI and Anthropic held these talks, even if they haven't finalized an agreement, demonstrates that safety concerns are reshaping how the world's most advanced AI labs operate. Whether the two companies ultimately sign the agreement remains to be seen, but the negotiations themselves signal a shift toward greater transparency and mutual accountability in an industry that has historically operated with significant secrecy around model development and safety testing.