Anthropic has decided to block OpenAI’s access to the APIs of its models known as Claude, citing a clear breach of the agreed commercial usage terms between the two parties. According to a report by Wired magazine, OpenAI exploited Claude models in internal tests and comparisons involving programming performance, content writing, and safety evaluations, which violates clauses prohibiting the use of Claude in developing competing products. These tests were reportedly part of OpenAI’s preparations to launch its upcoming GPT-5 version. A spokesperson for Anthropic stated that their technical tools were used during the pre-launch phase of GPT-5, clarifying that this “does not fall under permitted uses.” The company confirmed it will allow limited use of the models only for safety tests and academic assessments, while continuing to ban any integration within commercial tools or products.

OpenAI described the decision as “disappointing,” asserting that their actions fall within industry-standard practices. The company’s spokesperson added that they respect Anthropic’s decision despite OpenAI continuing to provide API access to competitors, including Anthropic itself. The decision was not entirely surprising, as Anthropic officials had previously expressed reservations about cooperating with companies potentially acquired by OpenAI. Previously, Jared Kaplan, Chief Scientist at Anthropic, said that selling Claude to a competitor like OpenAI would be “strange” and strategically illogical. These developments come amid unprecedented competition in the AI industry, with major companies like Anthropic, OpenAI, Google, and Meta racing to develop advanced language models, raising questions about fair use boundaries and transparency in data and algorithm sharing among major players.