Anthropic Withdraws OpenAI’s Access to Claude Models

A report from Wired reveals that Anthropic has discontinued OpenAI’s access to its Claude AI models.

Wired’s sources indicate that OpenAI was integrating Claude into its internal tools, allowing the company to evaluate Claude’s performance in areas like coding, writing, and safety against its own models.

TechCrunch has reached out to Anthropic and OpenAI for their responses. An Anthropic representative told Wired, “OpenAI’s technical team were also utilizing our coding tools prior to the release of GPT-5,” describing this as “a direct violation of our terms of service.” (Anthropic’s commercial agreements ban the use of Claude for developing competing services.)

Nonetheless, Anthropic stated it would continue to allow OpenAI access for “benchmarking and safety assessments.”

In reply, an OpenAI spokesperson noted, “While we acknowledge Anthropic’s decision to revoke our API access, it’s disappointing since our API remains available to them.”

Anthropic executives had previously shown hesitancy in granting access to competitors. Chief Science Officer Jared Kaplan defended the decision to cut ties with Windsurf (rumored as an acquisition target for OpenAI, which ultimately was acquired by Cognition) by stating, “It would be strange for us to sell Claude to OpenAI.”