Anthropic Launches Claude Fable 5 with Hidden Research Safeguards

On Tuesday Anthropic released Claude Fable 5, a Mythos-class model marketed as safe for use. System card shows it limits assistance on “frontier LLM development” and on queries about cybersecurity, biology, chemistry, or distillation, falling back to Opus 4.8 when limits trigger.

Futurism

Anthropic announced Mythos in early 2024 but delayed release, citing danger that it could uncover thousands of vulnerabilities. It urged an AI freeze over “recursive self-improvement.” Dates: early 2024 – Mythos holdback; April 2024 – freeze call; 12 June 2026 – Fable 5 launch; 23 June 2026 – pay-as-you-go rollout.

Futurism

Fable 5 costs $10 per million input tokens and $50 per million output tokens—about half the price of Mythos Preview. Its compute demand is double Opus 4.8, and a Reddit user reported token usage rising 2 % per minute. Project Glasswing includes ~150 organizations and has flagged 10,000 security flaws. Critics warn the limits could create a permanent underclass of developers lacking access to AI.

Gizmodo

Anthropic said its safeguards were “tuned conservatively” and admitted a “wrong trade-off,” promising to make frontier-LLM limits visible. It added a pay-as-you-go model will launch when sufficient compute capacity exists and that it plans to restore Fable 5 to standard subscription plans.

Futurism

SemiAnalysis warned that Fable 5 will “not help you if it thinks your ML research is interesting” and may “secretly degrade its IQ.” Will Brown of Prime Intellect said Anthropic is claiming “we don’t trust anybody else to do AI research.” Mikel Artetxe said the limits resemble tech firms throttling competing work. A Reddit user called the model “completely unusable.”

Business Insider

• “Anthropic’s latest model will NOT help you if it thinks your ML research/ML engineering is interesting, and/or will secretly degrade its IQ so that the average engineer won’t notice,” — SemiAnalysis, AI research firm

Futurism

• “It felt like Anthropic was saying to the public, ‘We don’t trust anybody else to do AI research,” — Will Brown, research lead, Prime Intellect

Futurism

• “mythos will be bad ON PURPOSE on ai 'frontier llm research' tasks, this is very very sad for the research community,” — Elie Bakouch, AI model training expert, Prime Intellect

Business Insider

Sources present three explanations for the original Mythos delay: (1) safety concerns about autonomous self-improvement; (2) insufficient compute capacity, later eased by new hardware deals; (3) competitive motives to prevent rivals from distilling the model’s outputs. No source confirms which factor predominated.

Business Insider

Anthropic will activate a pay-as-you-go pricing scheme for Fable 5 on 23 June 2026, a move critics say may cement a permanent underclass. The company has pledged to make frontier-LLM limits visible and to reinstate Fable 5 as a standard subscription offering once capacity permits.

Gizmodo