Anthropic retreats on a policy that would secretly restrict competitors from using a fresh AI model, Claude Fable 5, to develop other AI models. The company reversed course after the move was met with fierce opposition from the artificial intelligence research community.
“We are changing Fable 5’s protections for pioneering LLM development to be visible,” Anthropic said in a statement to WIRED. “We made a bad compromise and we apologize for not getting the balance right.”
Earlier this week, Anthropic released the Claude Fable 5, a version of its latest AI model with additional safety rails designed to prevent misuse. Some of the safeguards Anthropic opted for weren’t surprising: The company said it would redirect users asking questions about cybersecurity, biology or chemistry to a less productive AI model to reduce the risk of someone using advanced AI to launch a cyberattack or build a biological weapon.
However, for researchers trying to utilize Claude Fable 5 to pioneer artificial intelligence, Anthropic offered a different approach. The company intentionally reduced the model’s performance in a way that was hidden to the user. This move would effectively sabotage researchers attempting to utilize Claude to train competing artificial intelligence models, something Anthropic expressly prohibits in its operations. terms of service.
Anthropic now says it is changing course and that Claude Fable 5’s protections regarding AI development will be noticeable to users. If the company suspects that a user is trying to utilize Claude to build a powerful AI, it will notify the company to either deny the request or redirect the user to a less capable model.
Anthropic changed this policy after backlash from the artificial intelligence research community. Anthropic has already taken steps to restrict competitors from using Claude to create closed- and open-source AI models, but critics say that quietly degrading the model’s performance for some users has gone a step too far. The Claude coding agent has become a favorite tool of developers, including those working on open-source AI research projects, and researchers tell WIRED that the company’s latest policy could lead to a troubling future in which only a handful of leading AI labs will be able to conduct advanced AI research.
Dean Ball, senior fellow at the Foundation for American Innovation and former White House adviser on artificial intelligence, wrote in: post to X that “the demeaning results of machine learning research *without informing the user* are shockingly hostile and terrible.” He continued in another post that the “covert sabotage” policy undermines Anthropic’s overall position because it limits AI researchers from collaborating on AI security.
“I felt like Anthropic was saying to the public, ‘We don’t trust anyone else to do AI research. We’re the only ones who need to do AI research,'” says Will Brown, research manager at open-source AI startup Prime Intellect. “I feel like they’re starting to drag the ladder behind them.”
Brown said the policy would also leave developers unaware whether they were violating Anthropic’s policies because the company would not notify them when the protections were triggered. He added that the restrictions could have far-reaching consequences. For example, he pointed to a growing ecosystem of third-party rating companies that test pioneering models for safety, performance and reliability – work that could be made more tough if Anthropic quietly downgraded its model.