OpenAI lost access to Claude's API. Photo: Bloomberg . |
Anthropic, the company behind the Claude language model, has cut off OpenAI's access to its API after discovering misuse that violates its commercial terms of service.
According to Wired , multiple sources confirmed that OpenAI used Claude's API to help develop and test GPT-5, the next major version of its language model scheduled for release in August. However, instead of using a regular chat interface, OpenAI allegedly accessed Anthropic's internal tools.
From there, they ran a series of tests to assess Claude's performance and ability in tasks like coding, content creation, and handling sensitive situations like self-harm and hate speech.
The source said OpenAI’s goal was to compare Claude with GPT-5 to refine the new model, thereby increasing competition in the AI field. However, this action violated Anthropic’s commercial terms of service, which state that companies cannot use Claude to develop competing AI products unless Anthropic gives them permission.
“Customers may not attempt to access the Services to build competing products or tools, including training new AI models or reselling the Services,” reads Anthropic’s terms.
In response, OpenAI said that peer-reviewing rival models is a common practice in the AI industry. However, the company expressed disappointment that its access was blocked, especially given that Anthropic still has access to OpenAI’s API. The company said that the access to Claude was for “safety and ethical review purposes only.”
This isn’t the first time Anthropic has been tough on access. In June, the company stopped providing APIs to Windsurf, a startup rumored to be in talks to sell to OpenAI itself. The deal later fell through, but Anthropic’s move reflects a clear commitment to protecting its core technology.
Anthropic also recently announced that it would be adjusting the price limit for using Claude starting in late August, in part to prevent account sharing and reselling.
Source: https://znews.vn/tin-xau-lien-tiep-den-voi-openai-post1574072.html
Comment (0)