Anthropic Walks Back Policy That Could Have ‘Destroyed’ AI Researchers Using Claude

Anthropic and backward on a strategy that would have been to undercut competitors using its new AI model, Claude Fable 5creating other types of AI. The company changed practices after the move after receiving significant feedback from a AI research group.

“We are changing the security of Fable 5 to improve the limit of LLM to be visible.” Anthropic said in a statement to WIRED. “We made the wrong tradeoff and apologize for the inconvenience.”

Anthropic released Claude Fable 5, its latest AI version with additional security features designed to prevent misuse, earlier this week. Some of the safeguards Anthropic proposed were surprising: The company said it would also switch users who ask questions about cybersecurity, biology, or chemistry to a lower-level version of AI to reduce the chance of someone using advanced AI to carry out a cyberattack or create a bioweapon.

But for researchers trying to use Claude Fable 5 for AI development, Anthropic has outlined an alternative. The Company may intentionally disparage the Model’s work in ways that are not visible to the User. The move would effectively harm researchers trying to use Claude to train competing AI models, which Anthropic strictly prohibits. word of mouth.

Anthropic now says that it is changing, and that the security of Claude Fable 5 in the development of AI will be visible to users. If the company suspects that a user is trying to use Claude to create a more capable AI, it will warn them that it will reject the request, or switch the user to a lower version.

Anthropic changed the policy after receiving severe backlash from the AI ​​research community. Anthropic has already taken the initiative discourage competitors from using Claude building closed and open AI models, but critics say that quietly discrediting the model’s performance to some users went too far. Claude’s code assistant has become a favorite tool among developers, including those working on open-source AI research, and researchers told WIRED that the company’s latest policy could lead to a troubled future in which a handful of leading AI labs can conduct advanced AI research.

Dean Ball, a senior fellow at the Foundation for American Innovation and a former White House adviser on AI, wrote post on X that “the despicable work on ML research *without telling the user* is ugly and ugly.” He continued in another post that the principle of “privacy-destruction” undermines the entire Anthropic concept, because it prevents AI researchers from engaging with AI security.

“It felt like Anthropic was saying to people, ‘We don’t trust anyone else to do AI research. We have to do AI research ourselves,'” says Will Brown, director of research at startup AI Prime Intellect. “It sounds like they’re starting to drag the ladder behind them.”

Brown said the policy would also have left developers in the dark if they were violating Anthropic’s rules, since the company wouldn’t have warned them when its security was triggered. He added that the restrictions would have had many consequences. For example, he pointed to a growing ecosystem of human testing companies that test edge models for safety, performance, and reliability — work that would have been thwarted if Anthropic had secretly destroyed its model.