Anthropic retreats on a policy that would covertly limit competitors from using its new AI model, The story of Claude 5develop other models of AI. The company changed after the move received strong opposition from the company A community of AI researchers.
“We’re changing Fable 5’s protection for LLM’s development of the boundaries to make them visible,” Anthropic said in a statement to WIRED. “We made the wrong trade and we apologize for not getting the balance right.”
Anthropic released Claude Fable 5, a version of its latest AI model with additional safeguards designed to prevent abuse, earlier this week. Some of Anthropic’s defenses were decidedly unsurprising: The company said it would redirect users who asked questions about cybersecurity, biology, or chemistry to a less powerful AI model to reduce the chance someone could use advanced AI to carry out cyberattacks or create biological weapons.
But for researchers trying to use Claude Fable 5 for frontier AI development, Anthropic outlined a different approach. The company would deliberately degrade the performance of the model in ways that were not visible to the user. The move would damage researchers trying to use Claude to train AI models, which Anthropic prohibits terms of service.
Anthropic now says it is changing, and that Claude Fable 5’s defenses for AI development will be visible to users. If the company suspects that a user is trying to use Claude to create a powerful AI, it will notify them that it is rejecting the request, or redirecting the user to a less powerful design.
Anthropic reversed the policy after receiving strong backlash from the AI research community. Anthropic has already taken action reduce competitors using Claude building closed and open-source AI models, but critics say that degrading the model’s functionality for some users went too far. Claude’s coding agent has become a favored tool among developers, including those working on open-source AI research projects, and researchers tell WIRED that the company’s latest policy could lead to a troubled future where only a few leading AI labs could conduct advanced AI research.
Dean Ball, a senior fellow at the Foundation for American Innovation and a former White House adviser on AI, wrote in post on X that “poor performance on ML research *without telling the user* is shockingly hateful and scary looking.” He continued with another post that the “secret sabotage” policy undermines Anthropic’s overall position, because it limits AI researchers from collaborating on AI security.
“It felt like Anthropic was telling the public, ‘We don’t trust anyone else to do AI research. We’re the only ones who should be doing AI research,'” says Will Brown, research lead at open source AI startup Prime Intellect. “It feels like they’re starting to pull the ladder behind them.”
Brown said the policy would also leave developers in the dark about whether they’re violating Anthropic’s rules, since the company wouldn’t alert them when its protections were triggered. He added that the restrictions could have many effects. As an example, he pointed to a growing ecosystem of third-party organizations that test edge models for security, performance, and reliability—work that could have been prevented if Anthropic had cracked down on its model.
