Article illustration 1

In a dramatic escalation of the AI arms race, Anthropic has revoked OpenAI’s API access to its Claude models, citing violations of its commercial terms of service. According to internal sources and company statements confirmed by WIRED, OpenAI was using Anthropic’s developer tools to benchmark Claude’s coding and safety capabilities against its own in-development models—specifically the rumored GPT-5, which is expected to launch with enhanced programming features.

Anthropic spokesperson Christopher Nulty stated: “Claude Code has become the go-to choice for coders everywhere and so it was no surprise to learn OpenAI's own technical staff were also using our coding tools ahead of the launch of GPT-5. Unfortunately, this is a direct violation of our terms of service.” The terms explicitly bar using the service to “build a competing product or service” or “reverse engineer” its models.

OpenAI had been leveraging Anthropic’s API—rather than the standard chat interface—to run comparative tests on Claude’s performance in areas like code generation, creative writing, and responses to high-risk prompts involving self-harm or exploitation. This practice, while common for benchmarking AI safety and progress, crossed into forbidden territory when tied to GPT-5’s development. OpenAI’s chief communications officer Hannah Wong countered: “It’s industry standard to evaluate other AI systems to benchmark progress and improve safety. While we respect Anthropic’s decision, it’s disappointing considering our API remains available to them.”

This move isn’t isolated. Anthropic recently restricted API access for AI startup Windsurf amid rumors of an OpenAI acquisition, with chief science officer Jared Kaplan remarking, “It would be odd for us to be selling Claude to OpenAI.” The tactic echoes broader industry patterns: Facebook cut off Twitter’s Vine in 2013 over competitive concerns, and Salesforce limited Slack API access for rivals last month. Such restrictions underscore how APIs—once open gateways for innovation—are increasingly weaponized in tech turf wars.

For developers, this clash signals tightening API governance that could stifle cross-platform experimentation. Anthropic’s new rate limits on Claude Code, announced just before the OpenAI cutoff, hint at a broader crackdown on terms-of-service violations. Yet the stakes transcend corporate sparring. As generative AI advances, transparent benchmarking is crucial for safety and performance—but when rivals become test subjects, ethical and competitive lines blur. The real cost? Progress itself, as once-collaborative ecosystems fragment into walled gardens.

Source: WIRED