Claude kündigt die Zusammenarbeit mit OpenAI, und der Wettbewerb im Bereich der KI-Programmierung eskaliert erneut.

Anthropic suddenly revoked OpenAI employees' access to Claude, claiming that the company violated its terms of service.

According to a report from WIRED, OpenAI's technicians were still using Claude Code just before the release of GPT-5.

That is to say, OpenAI used its competitor's Claude to assist in the development and security testing of its own GPT-5. Moreover, according to the actual tests released by Altman, GPT-5 has rewritten the coding rules in the programming field.

Doesn't this trigger Anthropic's API defense strategy?

Let's take a look at what's going on in this AI programming battle.

Dario Amodei Criticizes Altman and Zuckerberg

In an interview program, Dario Amodei, the CEO of Anthropic, talked about the issue of competition among AI companies.

Dario Amodei believes that positive research affecting AI development should come from the corporate level, such as more effective methods, release timing, and research.

Regarding OpenAI's operations in this regard, he said:

"The leader of a company must be a trustworthy and sincerely motivated person. No matter how you push the company forward technically, if you work for someone who is insincere, dishonest, and doesn't really want to make the world a better place, it won't succeed. You're just contributing to something bad."

In addition to criticizing OpenAI, Dario Amodei also took a swipe at Zuckerberg's crazy talent - poaching behavior.

He said that Zuckerberg's move is like throwing darts. Whoever gets hit gets a salary ten times higher than others, which is undoubtedly subjective and unfair.

He also believes that Meta's behavior is an attempt to buy something that can't be obtained. He is pessimistic about whether Meta can attract the most talented, passionate, and enthusiastic talents.

As for whether he himself would do such a thing.

He said that Anthropic's work structure is hierarchical. Employees are graded and paid according to their positions when they join, and there is no further negotiation on this.

We won't break our principle of fairness to compromise on this kind of salary design, nor will we let such things destroy the company culture.

The meaning is that if we offer sky - high salaries to certain individuals, it will be detrimental to internal stability, so we won't do it.

Some people also criticized Dario Amodei, asking if he wants to monopolize the market.

Dario Amodei responded in the program that we've never said that Anthropic is the only company building this technology and similar ideas.

He just believes that in the end, everyone will be a loser in malicious competition. The real principle of competition should be to set an example for the operation of the entire field, and that's what Anthropic is doing.

Details of AI Programming Competition

According to Anthropic's commercial terms of service, customers are prohibited from using the service to "build competitive products or services, including training competitive AI models" or "reverse - engineer or copy" the service.

OpenAI is unhappy about this.

Hannah Wong, OpenAI's chief communications officer, said: "Evaluating other AI systems to measure progress and improve safety is an industry norm. While we respect Anthropic's decision to cut off our API access, it's disappointing considering our API remains open to them."

According to a source, OpenAI is integrating Claude into its internal tools through special developer access (API) rather than using the regular chat interface. This allows the company to run tests to evaluate Claude's capabilities in programming and creative writing compared to its own AI models and to check Claude's responses to security - related prompts such as CSAM, self - harm, and defamation.

The source said that these results help OpenAI compare the behavior of its own models under similar conditions and make adjustments as needed.

This incident has also sparked a heated discussion among netizens on social media.

Some netizens expressed their views from a technical perspective:

If the accusation is that OpenAI is explicitly using Claude's code output for training, then it is very reasonable. If the accusation is that OpenAI's technicians are using it for general programming, which indirectly leads to the construction of a competing model, then it seems absurd.

A netizen who works at Anthropic said that they will continue to provide API access for security evaluation and benchmarking.

He further added that API access is not sold like consumer products. The terms of both Anthropic and OpenAI are standard and are reciprocal protections. People working at OpenAI can also use Claude or other AI models under their personal identities.

Some people condemned Anthropic, saying it was a moral issue. Of course, there are also netizens who spoke up for Anthropic.

However, cutting off a competitor's API access has always been one of the competition strategies of technology companies.

Just last month, Anthropic significantly reduced the access of another well - known AI programming startup, Windsurf, to the Claude model, forcing Windsurf to find another computing power provider in the short term to continue its operations.

Let's see what new developments will happen in the future in this competition among technology companies in AI programming.

Reference Links

[1]https://x.com/ns123abc/status/1951434007792529744

[2]https://www.wired.com/story/anthropic-revokes-openais-access-to-claude/

[3]https://www.youtube.com/watch?v=mYDSSRS-B5U

[4]https://x.com/ai_for_success/status/1951577562511192087

This article is from the WeChat official account "QbitAI", author: Yiran. It is published by 36Kr with permission.