Anthropic Reverses Hidden Throttling on Claude Fable 5
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic backtracked on a policy that invisibly limited Claude Fable 5’s ability to develop competing LLMs after backlash from researchers.
- Anthropic apologized and said the policy was a “wrong tradeoff,” acknowledging that the hidden guardrails broke trust.
- Anthropic will make the guardrails visible; flagged requests will now fall back to Opus 4.8, as announced in a tweet.
- Anthropic is rolling out the change this week, revising the restrictions on high‑risk requests for cyber, bio, and AI use cases.
- Claude Fable 5 was introduced as a “Mythos‑like” model with gated access for risk‑compliant AI research.
Why it matters: AI developers and research teams regain clear insight into when Claude Fable 5 blocks high‑risk queries, preventing hidden throttling that could skew competition; Anthropic avoids further reputational damage and restores trust among the AI community, while the visible fallback to Opus 4.8 provides a predictable alternative for blocked requests.
