Anthropic has reversed a policy that would have covertly limited competitors from using its latest AI model, Claude Fable 5, to develop other AI systems. The company announced the change in June 2026 after the policy drew sharp criticism from the AI research community.
When Anthropic released Claude Fable 5 earlier that week, it included safety guardrails designed to prevent misuse. Some were straightforward: users asking about cybersecurity, biology, or chemistry would be rerouted to a less capable model to reduce risks tied to cyberattacks or bioweapons. But for researchers working on frontier AI development, Anthropic had planned a different approach — deliberately and invisibly degrading the model’s performance without alerting the user.
The covert restriction would have effectively sabotaged researchers attempting to use Claude to train competing AI models, a use Anthropic already prohibits in its terms of service. Critics argued that quietly degrading performance crossed a line beyond simply enforcing those terms.
“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” Anthropic said in a statement to WIRED. “We made the wrong tradeoff and we apologize for not getting the balance right.” Going forward, the company says it will notify users when it is refusing a request or rerouting them to a less capable model.
Dean Ball, a senior fellow at the Foundation for American Innovation and a former White House AI advisor, called the original policy “shockingly hostile,” writing on X that the “secret sabotage” undermines AI safety collaboration. Will Brown, research lead at open-source AI startup Prime Intellect, said the policy would have left developers unaware of whether they were violating Anthropic’s rules, and warned it could have disrupted third-party firms that evaluate frontier models for safety and reliability.
Brown added that the restrictions could have concentrated advanced AI research among only a handful of leading labs. “It felt like Anthropic was saying to the public, ‘We don’t trust anybody else to do AI research,'” he told WIRED. “It feels a bit like they’re starting to pull the ladder up behind them.”
Source: WIRED