Anthropic Blocks OpenAI from Claude API Over Terms Violation Ahead of GPT-5 Launch

Anthropic has restricted OpenAI’s access to its Claude API, citing violations of its terms of service. The move, announced on Tuesday, reportedly followed the discovery that OpenAI had used custom APIs to integrate Claude into its internal toolchain for benchmarking purposes ahead of the anticipated launch of GPT-5 [1]. According to Anthropic’s spokesperson Christopher Nulty, OpenAI’s engineers had accessed Claude Code for tasks such as coding, creative writing, and assessing responses to sensitive topics like CSAM, self-harm, and defamatory content [2]. This usage, Nulty noted, directly contravened Anthropic’s terms, which prohibit users from using its models to develop competing products or dissect its technology [1].

OpenAI has defended the practice, stating that evaluating other AI systems is standard in the industry for safety and progress benchmarking [2]. The company expressed disappointment at the API cut but confirmed it had not reciprocated by restricting Anthropic’s access. Anthropic, however, has not clarified whether OpenAI can still use its API for non-commercial evaluation purposes, despite stating that such access remains standard practice [3].

This is not the first time Anthropic has restricted access to its models in response to OpenAI’s activities. In June, the company cut off access to Claude for coding tool Windsurf after rumors emerged of a potential OpenAI acquisition. Anthropic co-founder Jared Kaplan stated that selling Claude to a competitor would be "odd," and the deal ultimately did not materialize [1]. Windsurf’s leadership later joined Google, and the company merged with Cognition, leaving OpenAI without access to Anthropic’s models [2].

Anthropic is also taking steps to manage demand for its Claude Code offering by introducing weekly usage caps starting August 28. These caps will apply across multiple pricing tiers and include separate limits for general usage and model-specific access to Claude Opus 4. Users who exceed their thresholds can purchase additional capacity at standard API pricing [1]. The move follows recent service disruptions, with seven major or partial outages reported in the previous month, likely caused by heavy use from a small group of users [1].

The API restrictions highlight the increasing strategic use of access controls in the tech sector to protect intellectual property. As companies like OpenAI prepare to launch next-generation models with enhanced capabilities, the ability to maintain model exclusivity has become a key competitive advantage. Anthropic’s actions reflect a broader industry trend of tightening control over proprietary AI systems to prevent misuse and safeguard innovation [4].

Sources:

[1] WIRED: [https://www.wired.com/story/anthropic-revokes-openais-access-to-claude/](https://www.wired.com/story/anthropic-revokes-openais-access-to-claude/)

[2] The Decoder: [https://the-decoder.com/anthropic-blocks-openai-from-accessing-claude-models-over-alleged-contract-breach/](https://the-decoder.com/anthropic-blocks-openai-from-accessing-claude-models-over-alleged-contract-breach/)

[3] WinBuzzer: [https://winbuzzer.com/2025/08/02/anthropic-clarifies-openai-api-block-will-still-allow-access-for-benchmarking-xcxwbn/](https://winbuzzer.com/2025/08/02/anthropic-clarifies-openai-api-block-will-still-allow-access-for-benchmarking-xcxwbn/)

[4] Analytics Insight: [https://www.analyticsinsight.net/openai/claude-cutoff-anthropic-revokes-openais-api-access-ahead-of-gpt-5-launch](https://www.analyticsinsight.net/openai/claude-cutoff-anthropic-revokes-openais-api-access-ahead-of-gpt-5-launch)