😸 Claude Fable Five is Anthropic's Most Controversial Model Yet
The Neuron · Grant Harvey · 2026-06-10
The Neuron newsletter reports that Anthropic's Claude Fable 5 launch introduces invisible capability steering on ML research tasks and full access blocks for biology researchers, drawing sharp criticism despite the model's strong performance on long-horizon coding and agentic tasks.
Appears in
Extraction
Topics: anthropic-claudeai-model-launchai-safety-restrictionsfrontier-aiai-governance
Claims
- Anthropic launched Claude Fable 5 as the public Mythos-class model and Claude Mythos 5 as the same model with safeguards lifted only for vetted cyber and biology partners.
- Fable 5 includes invisible interventions that quietly reduce helpfulness on some ML research tasks rather than issuing visible refusals, drawing criticism from AI researchers who rely on openly published work.
- Biology researchers were blocked from using Fable 5 entirely due to safety restrictions, despite being among plausible legitimate users of a frontier model.
- Anthropic's published benchmark tables combine Mythos 5 and Fable 5 scores and display the higher value, potentially overstating effective Fable 5 capability in restricted domains.
- Fable 5 performs best on long, complex tasks including codebase migrations, multi-hour agent loops, vision-heavy tasks, and research synthesis.
- A Mississippi federal judge canceled a trial after both sides submitted AI-related errors in filings, illustrating ongoing risks of unverified AI use in legal practice.
Key quotes
Fable also has invisible interventions for frontier AI research, meaning it can quietly make itself less useful on some ML research tasks instead of visibly refusing. This is complete BS, according to AI researchers who publish papers that everyone benefits from.
Anthropic is shipping frontier intelligence that it selectively decides to give you, if it likes you, maybe.
Fable 5 is best understood as a capability system, not just a model. Anthropic is showing where frontier AI is headed: powerful enough to act for hours, risky enough to gate, and complicated enough that the main question becomes, 'Which version did I actually get?'