Anthropic says OpenAI staff used Claude Code ahead of GPT-5 launch and it will continue providing API access to OpenAI for benchmarking and safety evaluations
Anthropic says it has revoked OpenAI's access to the Claude API after ChatGPT's engineers were found using Claude's coding tools.
Context & Ripple Effects
Anthropic and OpenAI have been competitors since Claude’s launch, with the rivalry extending to coding capabilities; prior coverage described OpenAI strengthening ChatGPT’s coding work in response to Claude. The latest dispute turns a commercial API relationship into a point of competitive control.
The reported revocation follows Anthropic’s allegation of terms-of-service violations tied to model comparison. Anthropic’s stated decision to retain a narrow channel for benchmarking and safety evaluations distinguishes permitted evaluation from use of Claude Code in product development.
First-order effects
- OpenAI loses the reported Claude API access used by ChatGPT engineers for Claude Code, while retaining the access Anthropic says is necessary for benchmarking and safety evaluations.
- Anthropic is enforcing a more bounded relationship with a direct rival: comparative testing remains permitted, but access it characterizes as competitive product use does not.
Second-order effects
- The distinction makes API terms and access controls more consequential for frontier-model developers that both sell services and compete with their customers.
- The episode may push rivals to formalize how they separate benchmarking from development workflows, especially in coding, where competitive responses around coding capabilities have already been part of the rivalry.
Third-order effects
- If similar restrictions proliferate, cross-model evaluation could increasingly operate through negotiated, auditable access rather than broadly available commercial APIs.
- Frontier-model providers may treat API governance as both a safety mechanism and a competitive boundary, creating pressure for clearer norms around legitimate model comparison.
The trend: Frontier AI companies are moving from open-ended API distribution toward more tightly governed access that distinguishes evaluation from competitive product development.