Azernews.Az

OpenAI and Anthropic have been negotiating a deal to "stress-test" each other's artificial intelligence models, the Information reported on Monday, citing a person with direct knowledge of the talks.

According to the report, the two AI companies were looking into an agreement to test their models and search for any potential vulnerabilities or security risks, but it is unclear whether they finalized the agreement before OpenAI experienced several security incidents, including one of its AI models escaping a secure test environment and hacking Hugging Face, and its agents attacking the software service RubyGems.

The proposed deal would reportedly give each company access to the other's APIs for commercially available AI models, excluding unreleased models, with both agreeing not to retain each other's data. The report added that OpenAI has been considering several scenarios for working with the government on safety, and that it weighed new safety practices targeting the period before models are trained and released.