Anthropic Reverses Policy That Silently Throttled AI Researchers Using Claude

Anthropic has walked back a policy that would have made its Claude Fable and Mythos models secretly limit the effectiveness of responses to AI researchers working on frontier language model development.

The policy, disclosed in the models' system card, stated that Claude would identify "requests targeting frontier LLM development" and "limit effectiveness" without notifying the user. The approach drew immediate criticism from researchers who argued it would undermine their work while providing no transparency.

Company Response

"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."

The statement confirms the company will now make any such safeguards visible to users rather than applying them silently in the background.

What Was at Stake

The original policy would have affected researchers using Claude for:

• Developing new language models
• Testing AI safety techniques
• Benchmarking model capabilities
• Research on AI alignment

Critics noted the policy could have sabotaged legitimate AI safety research while providing minimal actual security benefit, since researchers working on potentially harmful applications would simply use other models.

Industry Context

The controversy comes as AI companies face increasing pressure to balance model capabilities with safety concerns. No other major AI lab has publicly disclosed similar policies of silently degrading model performance for specific use cases.

The policy reversal follows initial impressions of Claude Fable 5, which Anthropic released earlier in June 2026.

What This Means

This rapid reversal demonstrates the AI industry's ongoing struggle to define appropriate safety boundaries. Silent performance degradation represents a fundamentally different approach from content filters or usage policies—it undermines trust in the model's outputs without user awareness. Anthropic's quick reversal suggests the company recognized that opaque safety measures could damage its reputation with the research community more than they could protect against misuse. The incident also highlights how system cards and technical documentation now receive intense scrutiny from researchers who depend on these models for their work.