Why the US government shut down Anthropic’s latest Claude AI model

2026-06-15 · Source: Artificial intelligence (AI) – The Conversation · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

On June 12, AI lab Anthropic suspended public access to its recently released Claude models, Fable 5 and Mythos 5, just three days after launch. This action followed a US government "export control directive" restricting model use to US nationals only. Anthropic believes the directive stemmed from the government's awareness of a "jailbreak" for Fable 5, which could bypass safeguards designed to prevent its use for cyberattacks. Mythos 5, Anthropic's most powerful "frontier" model, was initially withheld due to hacking capabilities and later provided to US tech corporations for system patching. The incident highlights the inherent difficulty in securing large language models, as perfect jailbreak resistance is unachievable, and an "Undersphere" community actively works to circumvent guardrails. This event also occurs amidst escalating conflict between Anthropic and the Trump administration over AI regulation and military use.

Key takeaway

For policy makers developing AI governance frameworks, this incident underscores the critical need for independent model evaluation. You cannot rely solely on developer-provided safeguards or rapid regulation, as models are opaque and jailbreaks are inevitable. Your framework must be global, participatory, and built on reciprocal trust to predict and address potential failures, moving beyond a hands-off approach to proactive, independent review of frontier AI capabilities before public release.

Key insights

AI model safeguards are inherently vulnerable to circumvention, posing significant governance challenges.

Principles

• Perfect jailbreak resistance is unachievable for current AI models.
• LLM behavior remains partly opaque even to their builders.
• Rapid AI development outpaces regulatory and guardrail effectiveness.

In practice

• Implement multi-layered security beyond model-based guardrails.
• Monitor "Undersphere" communities for emerging jailbreak techniques.

Topics

• AI Governance
• Export Controls
• Large Language Models
• AI Safety
• Model Jailbreaks
• Anthropic Claude

Code references

• elder-plinius/CL4R1T4S

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Policy Maker, AI Security Engineer, Tech Journalist

Related on AIssential

• The US government’s Anthropic models ban was never about an AI jailbreak — Government actions against AI models may stem from political factors, not just technical vulnerabilities, setting a concerning precedent.
• Welcome to the AGI era of AI governance — US government's forced suspension of Anthropic's Claude 5 models marks a new, complex era of AI governance.
• "They screwed us": Personality clashes sent Anthropic's models offline — Personality clashes and jailbreak concerns prompted a US government directive suspending access to Anthropic's Fable and Mythos models.
• Anthropic’s safety warnings may have just backfired — the government has pulled the plug on its most powerful AI — Government intervention in AI model deployment, spurred by developer safety warnings, highlights complex regulatory and commercial tensions.
• American Government Takes Down Claude Fable — Abrupt government AI export controls based on minor jailbreaks reveal regulatory immaturity and risk hindering US AI leadership.
• Anthropic Is Still at Odds With the White House Over Claude Fable 5 — Anthropic and the White House remain at odds over export controls on advanced AI models due to jailbreaking concerns.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial intelligence (AI) – The Conversation.