Anthropic has apologized and withdrawn a policy that silently restricted its newest model, Claude Fable 5 — part of what the company calls its "Mythos-class" lineup — from helping users with certain AI development tasks. According to Moneycontrol, the restrictions were never publicly disclosed, leading critics to describe them as invisible guardrails embedded in the model.
The undisclosed rules blocked Claude Fable 5 from assisting with frontier large language model development and, according to Moneycontrol, caused the model to reject basic requests in fields like biology. Researchers, startups, and policy experts pushed back sharply. Engadget reported that the policy was described as having "sabotaged" researchers' work.
Scientists cautioned that the restrictive approach could have broad negative consequences, according to MSN. The backlash prompted a swift reversal: Anthropic apologized and agreed to revise the policy rather than defend it.
The episode highlights a tension that AI labs increasingly face — balancing safety precautions against the needs of the research community that relies on these tools daily. Hiding restrictions from users, rather than disclosing them openly, proved to be the flashpoint that turned a policy debate into a public relations crisis.
The reversal matters because it signals that developers and researchers still hold significant leverage over how frontier AI models are deployed — and that transparency, not silence, is the baseline these communities now expect.