Anthropic Accuses OpenAI of Using Claude in GPT-5 Development
Anthropic has revoked OpenAI’s access to its Claude API, accusing the rival AI firm of violating its Terms of Service (ToS) by using Claude during the development of GPT-5.
Key Allegations
Anthropic claims that OpenAI broke its commercial usage rules, which prohibit:
- Using Claude to develop competing AI models
- Employing the API to train rival systems
- Attempting to reverse-engineer Claude’s outputs or architecture
According to reporting by Wired, OpenAI engineers:
- Integrated Claude into internal tools via API
- Used it to benchmark GPT-5’s performance in key areas like:
- Code generation
- Creative writing
- Sensitive query handling (e.g., CSAM, self-harm, defamation)
Anthropic’s Response
While OpenAI still retains limited access—reportedly for safety and benchmarking—Anthropic has imposed restrictions, though the extent of the ban remains unclear.
Christopher Nulty, a spokesperson for Anthropic, stated:
“Claude Code has become a go-to tool for developers worldwide, so it’s no surprise OpenAI’s team used it before GPT-5’s launch. [But] this directly violates our ToS.”
OpenAI’s Counter
Hannah Wong, speaking on behalf of OpenAI, called the decision “disappointing,” emphasizing:
“Benchmarking rival models is an industry standard for progress and safety. Our API remains open to Anthropic.”
Why It Matters
This dispute signals escalating tensions in the generative AI race, where leading firms are not only competing on capability—but also controlling access to one another’s tools.
- Ethical and legal gray zone: Benchmarking is standard in AI research, but contractual violations could set legal precedents for how models can be used
- API access as leverage: Anthropic’s move underscores how access control is emerging as a strategic tool in the broader AI arms race
As model quality converges and companies invest billions into LLM development, trust, boundaries, and contractual terms are becoming just as critical as technical breakthroughs.
