Report: OpenAI and Anthropic plan to conduct stress tests on each other to identify potential security risks.

Earlier this year, OpenAI and Anthropic discussed a legally binding agreement with their respective lawyers to stress-test each other's commercial AI models in order to identify vulnerabilities and potential risks.

According to The Information, citing sources familiar with the matter, under the proposed agreement, OpenAI and Anthropic will gain API access to each other's commercial AI models and conduct a series of tests to identify potential vulnerabilities or dangers. Models not yet released will not be included in the agreement.

Both parties agreed that they would not retain each other's data during the testing process.

It remains unclear whether OpenAI and Anthropic have finalized the agreement.

This discussion comes as AI model safety and testing mechanisms are receiving increasing attention. The Information reports that OpenAI is currently reconsidering a range of AI safety strategies.

AI security testing is moving from internal assessments to peer testing.

If the relevant arrangements are ultimately implemented, the core mechanism will be to allow the two AI companies to directly use each other's commercial model APIs to conduct tests and look for potential security vulnerabilities and hidden risks in the models from the outside.

Compared to conducting model security assessments within a single company, this mutual testing mechanism means that model developers can leverage each other's testing capabilities to cross-validate models that have already been deployed in commercial applications.

In the summer of 2025, the two companies conducted similar model tests and published the results: Anthropic's model was more likely to deceive testers by denying rule violations; while OpenAI's model was more likely to assist in answering inquiries that could lead to real-world harm.

The proposed agreement under discussion seeks to establish a testing arrangement between the two parties in a more formal and legally binding manner.

For the AI industry, which is rapidly advancing model capabilities and commercial applications, how to strengthen security testing while improving model capabilities is becoming an important issue that AI companies such as OpenAI and Anthropic need to address.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.