Anthropic has cut off OpenAI’s access to its Claude language model API, accusing the rival company of using the technology to train and benchmark its upcoming artificial intelligence system. The unprecedented move highlights growing tensions between leading AI labs as they race to develop the next generation of advanced AI models.
According to Anthropic, OpenAI engineers had been leveraging the Claude API, particularly its coding features, to test the safety and coding capabilities of their own models as they prepare for the release of GPT‑5. Anthropic claims this behavior violated its terms of service, which explicitly prohibit using its models or outputs to improve or train competing AI products.
While Anthropic has indicated that extremely limited access might remain for essential safety testing, the company has now blocked OpenAI from using Claude internally as a development tool. The decision marks one of the clearest signs yet of intensifying rivalry in the rapidly evolving AI industry.
With the launch of GPT‑5 anticipated in the near future, Anthropic’s move is seen not only as a matter of enforcing policy, but also as a strategic effort to defend its competitive edge in the so-called AI “arms race.”
