Anthropic Blocks OpenAI's Access to Claude API

Anthropic Revokes OpenAI's API Privileges

Anthropic has terminated OpenAI's access to its Claude API, citing violations of terms of service. Multiple sources confirm OpenAI lost access this week after using Claude for benchmarking against its upcoming GPT-5 model.

Terms of Service Breach

According to Anthropic's commercial terms, customers cannot use their services to "build a competing product or service" or "reverse engineer" their technology. Anthropic spokesperson Christopher Nulty stated OpenAI's technical staff had been using Claude's coding tools during GPT-5's development phase.

OpenAI's Benchmarking Practices

OpenAI utilized Claude's API for internal testing, evaluating capabilities in coding, creative writing, and safety responses. This allowed comparison of behavior across categories like CSAM detection and defamation prevention.

Industry Reactions

OpenAI's communications chief Hannah Wong called the revocation "disappointing," noting such evaluations represent industry standards for safety benchmarking. Anthropic maintains it will permit API access for standard benchmarking purposes.

Historical Precedents

This follows similar industry actions including Facebook's restriction of Vine and Salesforce limiting Slack API access. Anthropic previously revoked access from AI startup Windsurf during acquisition talks with OpenAI.