Anthropic Revokes OpenAI API Access Over GPT-5 Development

On July 29, 2025, Anthropic revoked OpenAI's API access to its Claude family of AI models. The company discovered that OpenAI technical staff had integrated Claude into internal tools to refine the upcoming GPT-5 model, an action Anthropic described as a direct violation of its terms of service which prohibit using the technology to build competing products or train competing models.

OpenAI defended the activity as an industry standard for benchmarking and safety evaluations. The company reported using the API to compare model behavior in coding, creative writing, and safety prompts, including tests on sensitive content such as defamation and self-harm. OpenAI expressed disappointment that its access was severed while its own API remains available to Anthropic.

Anthropic later clarified that it would reinstate OpenAI's access specifically for standard benchmarking and safety evaluations, though it maintained the ban on usage for model development. This conflict follows a pattern of restrictive behavior by Anthropic, which previously cut off API access for the coding startup Windsurf amid rumors of an acquisition by OpenAI.