Anthropic revoked OpenAI's API access to Claude, citing ToS violations; sources: OpenAI's use of the API let it compare its models' behavior against Claude's
OpenAI lost access to the Claude API this week after Anthropic claimed the company was violating its terms of service.
WiredKylie Robison
Context & Ripple Effects
The immediate follow-up clarified that Anthropic distinguished alleged use of Claude Code before GPT-5 from permitted benchmarking and safety work, saying it would retain API access for those narrower purposes. That turns an access cutoff into a dispute over what kinds of cross-model evaluation a rival may conduct.
Later, Anthropic made similar ToS allegations against other model makers over alleged high-volume prompting and distillation, suggesting the enforcement dispute expanded beyond OpenAI. The episode makes API terms a practical boundary around competitive intelligence, not merely a billing contract.
First-order effects
OpenAI loses the Claude API access Anthropic says was used to compare model behavior, disrupting that route for internal evaluation.
Anthropic immediately gains discretion to separate permitted benchmarking and safety evaluations from uses it considers competitive misuse, as outlined in its narrower API-access exception.
Second-order effects
Rival model developers using one another’s APIs face greater compliance risk: evaluation workflows may need clearer purpose limits, logging, and approval controls.
API providers have an incentive to make access tiers and terms more explicit when customers are also direct model competitors, particularly around model comparison and distillation-adjacent activity.
Third-order effects
If enforcement becomes routine, frontier-model APIs may evolve from broadly available developer utilities into selectively governed channels, with competitors receiving more constrained access than ordinary customers.
The dividing line between legitimate safety benchmarking and competitive model extraction could become a recurring source of contractual conflict; the available coverage establishes allegations and policy responses, not a settled industry standard.
The trend: This is one data point in the balkanization of frontier-model access, as providers treat API usage controls as a competitive and security boundary.
Love watching OpenAI and Anthropic accuse each other of ToS violations like two data hoarders pretending they didn't both loot Wikipedia (and everyone else) at 3am. [embedded post]
My guess (with no inside info from either side) is this is more of an anti-OAI thing than an Anthropic always enforcing this thing. Tons of competitors certainly use Claude Code.
@kyliebytes a lot of ppl are gonna assume this is purely for the sake of being anti-competitive because they dont think critically enough about what openai actually does with full access
IANAL but I would have thought that clause meant “don't distill from our model,” not “don't use the product in a normal product-y way.” It is hard to enforce either one without limiting totally benign use cases like safety benchmarking, so it is a bit surprising to see this.
very interesting. it's something of an open secret that frontier AI labs use each others' chatbots to improve their own... despite terms of service pretty clearly banning this
“as part of our effort to democratize the fruits of ai and ensure the singularity is helpful, harmless, and honest, we have decided to deny others working on the singularity the ability to chat with our helpful, harmless, and honest model”
Scoop: Anthropic revoked OpenAI's API access to its models on Tuesday, multiple sources familiar with the matter tell WIRED. OpenAI was informed that its access was cut off due to violating the terms of service. [image]
As Anthropic co-founder Jared Kaplan told me in June: “I think it would be odd for us to be selling Claude to OpenAI.” Also, very interesting choice for Anthropic's spokesperson to mention GPT-5's upcoming launch in this statement.
OpenAI comms says “while we respect Anthropic's decision to cut off our API access, it's disappointing considering our API remains available to them.” https://www.wired.com/...
@aidan_mclau This take is sad to see but you might not have full context. We cut OpenAl's access for violating our APl terms and for the heavy usage of Claude Code among OAI tech staff. We're going to continue providing API access for safety evals and benchmarking. That's importa…