Anthropic Withdraws OpenAI’s Access to Claude Models

According to a report by Wired, Anthropic has cut off OpenAI’s access to its Claude AI models.

Wired’s sources indicate that OpenAI was using Claude within its internal systems to evaluate its performance in areas like coding, writing, and safety compared to its own models.

TechCrunch reached out to both Anthropic and OpenAI for comments. An Anthropic spokesperson told Wired, “OpenAI’s technical team was using our coding tools prior to the release of GPT-5,” which was described as “a direct breach of our terms of service.” (Anthropic’s agreements prohibit the use of Claude for creating competing products.)

However, Anthropic stated it would still allow OpenAI access for “benchmarking and safety assessments.”

In a response, an OpenAI spokesperson said, “While we acknowledge Anthropic’s decision to revoke our API access, it’s disheartening, considering that our API remains available to them.”

Anthropic executives have previously shown hesitation in granting access to competitors. Chief Science Officer Jared Kaplan defended the move to cut ties with Windsurf (reportedly a target for OpenAI’s acquisition, which eventually went to Cognition) by stating, “It would be unusual for us to sell Claude to OpenAI.”