Anthropic Reverses Secret AI Research Restrictions After Backlash

AI laboratory Anthropic apologized on June 11, 2026, and announced revisions to the safety safeguards of its Claude Fable 5 model following intense criticism from the research community. The controversy began after the June 9 release of the model, when a 319-page system card revealed that Anthropic had implemented invisible interventions that silently degraded responses to requests involving cutting-edge AI development, such as training infrastructure and pretraining pipelines.

Researchers and policy experts, including representatives from the Foundation for American Innovation, described the policy as secret sabotage. Critics argued that modifying weights or prompts without transparent communication set a dangerous precedent for AI manipulation and fostered monopolistic behavior by concentrating capabilities.

Anthropic initially defended the measures, claiming they were necessary to prevent foreign adversaries and competitors—specifically citing Chinese firms like DeepSeek, Moonshot AI, and MiniMax—from using the model to build rival systems through distillation. The company estimated that only 0.03% of traffic was affected. However, following the outcry, Anthropic committed to making these safeguards visible. Flagged requests will now either provide a specific reason for refusal via the API or visibly fall back to the less powerful Opus 4.8 model.