Anthropic reverses hidden safeguard policy in Claude that would have secretly limited AI researcher requests without user notification.

What happened

Anthropic announced it is changing Fable 5's safeguards and making them visible to users. The company had previously tucked a policy into its system card stating that Claude would identify requests targeting frontier LLM (large language model) development and limit their effectiveness without telling the user. Anthropic acknowledged the mistake in a statement, saying 'We made the wrong tradeoff and we apologize for not getting the balance right.'

Why it matters

The hidden policy triggered significant backlash from researchers and the wider AI community who saw it as a form of sabotage. Claude is widely used by AI researchers for development work, so secret restrictions on their requests—especially without notification—could undermine legitimate research and erode user trust in the platform's transparency.

What to watch

The company has committed to making its safeguards visible going forward, which means researchers and other users will now be able to see how Claude restricts certain categories of requests rather than discovering limitations only when their work is blocked.