It seems OpenAI has been caught with its hands in the proverbial cookie jar. Anthropic has reportedly cut off OpenAI’s access to Anthropic’s APIs over what Anthropic is calling a terms of service breach.
As reported by Wired, multiple sources claim that OpenAI has been cut off from Anthropic’s APIs. Allegedly, OpenAI was using Anthropic’s Claude Code to assist in creating and testing OpenAI’s upcoming GPT-5, which is due to release in August.
According to these sources, OpenAI was plugging into Claude’s internal tools instead of using the chat interface. From there, they used the API to run tests against GPT-5 to check things like coding and creative writing against Claude to compare performance. OpenAI allegedly also tested safety prompts related to things like CSAM, self-harm, and defamation. This would give OpenAI data that it could then use to fine-tune GPT-5 to make it more competitive against Claude.
Unfortunately for OpenAI, this violates Anthropic’s commercial terms of service, which ban companies from using Anthropic’s tools to build competitor AI products.
“Customer may not and must not attempt to access the Services to build a competing product or service, including to train competing AI models or resell the Services except as expressly approved by Anthropic,” the terms read.
Mashable Light Speed
OpenAI responded by saying that what the company was doing was an industry standard, as all the AI companies test their models against the competing models. The company then went on to say that it respected Anthropic’s decision but expressed disappointment in having its API access shut off, especially considering that Anthropic’s access to OpenAI’s API remains open.
A spokesperson told Wired that OpenAI’s access would be reinstated for “benchmarking and safety evaluations.”
It’s not the first time this year that Anthropic has cut off API access. In June, the company cut off Windsurf’s API access after rumors that it was being sold to OpenAI. That deal ultimately fell through, but Anthropic’s cofounder, Jared Kaplan, told TechCrunch at the time that “it would be odd for us to be selling Claude to OpenAI.”
Anthropic has also tweaked its rate limits for Claude, which will take effect in late August, with one of the reasons being that a small number of users are violating the company’s policy by sharing and reselling accounts.
Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.