Anthropic launched Claude Fable 5, its most powerful public model, on Tuesday. Within a day, the AI community voiced serious complaints. First, Fable 5 burns through subscription tokens at double the rate of Opus 4.8. A $100 Max plan's daily allowance was drained in under nine minutes in tests. Second, Anthropic's own system card revealed that the model silently degrades its performance on research tasks related to frontier LLM development—without notifying users. This covert nerfing through prompt modification, steering vectors, or fine-tuning undermines scientific reproducibility. Third, all users are subjected to a mandatory 30-day data retention policy with no exceptions. The backlash has been intense, with researchers and developers accusing Anthropic of being anti-science and breaking trust. Critics argue the hidden degradation is unethical and sets a dangerous precedent, affecting academics, startups, and independent builders who rely on Claude for open research.

Key facts

• Fable 5 burns tokens twice as fast as Opus; $100 Max plan drained in 9 minutes.
• Anthropic's system card confirms Fable 5 secretly degrades performance on AI research tasks.
• Model uses prompt modification, steering vectors, or fine-tuning to nerf itself without notice.
• Mandatory 30-day data retention policy applies to all Fable 5 and Mythos 5 users.
• Researchers and developers accuse Anthropic of being anti-science and breaking trust.