Anthropic reverses silent nerfing policy for Claude Fable 5 AI development

Anthropic has walked back its controversial policy of silently downgrading Claude Fable 5's capabilities for users attempting to develop highly capable AI systems. In a statement to WIRED, the company said, "We’re changing Fable 5’s safeguards for frontier LLM development to make them visible. We made the wrong tradeoff and we apologize for not getting the balance right." The reversal means that if Anthropic suspects a user is trying to use Claude to build another powerful AI, it will now clearly alert them — either that it's refusing the request or that it's rerouting them to a less capable model. This ends the practice of "silent nerfing," which had sparked backlash from developers who encountered unexplained failures and degraded performance.

The decision comes amid growing scrutiny of safety measures that limit model misuse without transparency. Anthropic's original approach aimed to prevent users from leveraging Claude Fable 5 to create competing frontier models, but the lack of feedback frustrated legitimate researchers and developers. By making safeguards visible, Anthropic hopes to rebuild trust while still maintaining guardrails. The change applies to all users interacting with Claude Fable 5, with immediate effect. Going forward, developers will receive explicit notifications when their requests trigger a safety filter, allowing them to adjust their approach or pivot to alternative tools. The episode highlights the delicate balance between preventing misuse and supporting open AI research.