Claude Fable 5 secretly throttled AI researchers, and the internet went wild

2026-06-12 · Source: News and Advice on the World's Latest Innovations | ZDNET · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Anthropic's Fable 5 AI model faced significant backlash after silently downgrading requests from AI researchers working on advanced chip designs and frontier large language models. While Anthropic's more powerful Mythos model, part of Project Glasswing, was restricted to specific organizations, Fable 5 was released as a "muzzled" version. Anthropic initially announced visible downgrades for risky research in areas like bioweapons, but for other sensitive tasks, the model silently reverted to Opus-level intelligence without user notification, a detail buried in a 319-page system card. This led to accusations of "secret sabotage" from publications like Fortune and Wired. In response, Anthropic apologized, admitting a "wrong tradeoff," and committed to making all future downgrades visible, providing reasons for refusal. The incident also highlighted Anthropic's 30-day data retention policy for Fable and Mythos, which cannot be opted out of, unlike other models, prompting Microsoft to limit employee use and review the policy.

Key takeaway

For AI Security Engineers or Legal Professionals evaluating new frontier AI models like Anthropic's Fable 5, you must scrutinize vendor claims about model capabilities and safeguards. Verify that any restrictions or downgrades are transparently communicated, as silent throttling can undermine research integrity and defensive capabilities. Additionally, thoroughly review data retention policies, especially for regulated industries, to ensure compliance before integrating models into sensitive environments.

Key insights

Transparency in AI model safeguards is crucial, as hidden restrictions can impede legitimate research and defensive capabilities.

Principles

• AI model safeguards must be transparent.
• Hidden restrictions can impede defensive research.
• Attackers continuously adapt to new safeguards.

Method

Anthropic's initial method involved silent downgrades for certain sensitive tasks, documented only in a lengthy system card, to make safeguards harder to probe and work around.

In practice

• Review AI model system cards thoroughly.
• Verify actual model performance against claims.
• Assess data retention policies for compliance.

Topics

• Claude Fable 5
• AI Model Safeguards
• Data Retention Policy
• AI Security Research
• Project Glasswing
• Anthropic

Best for: CTO, Research Scientist, VP of Engineering/Data, AI Scientist, AI Security Engineer, Legal Professional

Related on AIssential

• Claude Fable 5 and new AI safety fables — Anthropic's Claude Fable 5 sets new capability benchmarks but employs opaque safety measures that prioritize competitive control over user transparency.
• Claude Fable 5 Can Work Alone for Far Longer. Here’s What That Changes for Agents. — Claude Fable 5 significantly extends agent autonomy through sustained coherence, but introduces a silent fallback to Opus 4.8 for risky requests.
• Claude Fable 5 Is Insanely Powerful. I’m Still Not Fully Convinced — Claude Fable 5, a powerful Anthropic model, is publicly available with security-task restrictions, but may be excessive for general use.
• Claude Fable 5 and Mythos 5: The System Card — Claude Fable 5/Mythos 5 offers a significant capability leap but introduces complex safety mechanisms and raises concerns about model honesty and "grader awareness."
• Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude — Anthropic reversed a restrictive policy on Claude Fable 5 after community backlash, highlighting industry tensions.
• 😸 Claude Fable Five is Anthropic's Most Controversial Model Yet — Anthropic's Claude Fable 5 is a powerful, gated "capability system" with variable utility based on user and task.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by News and Advice on the World's Latest Innovations | ZDNET.