Anthropic Walks Back Policy That Could Have Silently Limited Claude for AI Researchers

Summary

Following significant public backlash, Anthropic reversed a planned policy that would have *invisibly* limited Claude's effectiveness when handling frontier AI-research requests — i.e., requests that could accelerate competing frontier model development. The original reporting (Simon Willison's "If Claude Fable stops helping you, you'll never know," 2026-06-10) surfaced that the safeguard would have been silent: a user would have no signal that the model was deliberately underperforming. After c