The US government may be asking Anthropic the impossible by demanding unhackable LLMs

2026-06-15 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Intermediate, quick

Summary

The US government is in conflict with Anthropic, accusing the AI company of disregarding Trump's cyber executive order by releasing its Fable 5 model without explicit approval from a designated clearinghouse. Government officials, reportedly tipped by Amazon and other tech companies, claim Anthropic knew about a potential "jailbreak" in Fable 5. This accusation highlights the government's perceived lack of understanding of AI security, as experts like OpenAI have stated that prompt injection, a related hacking method, may never be fully solved, implying no LLM is truly "unhackable." Meanwhile, over 100 security experts and tech executives, including Alex Stamos and Rachel Tobac, have published an open letter defending Anthropic. They argue that Fable 5, along with models like GPT-5.5, Opus, Sonnet, and Chinese Kimi 2.7, are valuable tools for finding security flaws and that export controls would strip defenders of essential resources, especially as Chinese open-weight models are rapidly catching up. Anthropic CEO Dario Amodei previously warned in 2023 that jailbreaks could be "life or death."

Key takeaway

For policymakers considering AI export controls, you must recognize that demanding "unhackable" frontier AI models is currently an impossible standard. Your focus should shift from preventing all vulnerabilities to rapidly developing and deploying countermeasures, as even leading AI companies acknowledge inherent security risks. Restricting access to advanced models like Fable 5 could inadvertently disadvantage your own cybersecurity efforts, especially given the rapid progress of international competitors.

Key insights

The US government's demand for "unhackable" LLMs reveals a fundamental misunderstanding of current AI security limitations.

Principles

• All AI models are susceptible to hacking.
• Prompt injection may be an unsolvable problem.
• Export controls on AI tools hinder cybersecurity defense.

In practice

• Use advanced LLMs like Fable 5 to identify software vulnerabilities.
• Implement countermeasures quickly when breaches occur.

Topics

• AI Governance
• LLM Security
• Export Controls
• Anthropic Fable 5
• Prompt Injection
• Cybersecurity Tools

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

Related on AIssential

• [AINews] Fable and Mythos officially too dangerous to release — Geopolitical risks and "model sovereignty" are now explicit concerns for reliance on closed frontier AI APIs.
• 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.
• US government forces Anthropic to disable Claude Fable 5 and Mythos 5 for all customers worldwide — Government-mandated global disabling of Anthropic's Fable 5 and Mythos 5 due to a suspected jailbreak sets a contentious industry precedent.
• Politico folds E&E News ahead of broader energy push — Anthropic released Fable 5, a powerful AI model with robust safeguards, making advanced capabilities publicly accessible while mitigating misuse risks.
• AI, Government Control, and the Beginning of a New Technological Era — Government intervention on frontier AI models signals their reclassification as strategic, dual-use infrastructure.
• Why did the US government ban Claude Fable 5? — The US government banned Claude Fable 5 due to a perceived jailbreaking vulnerability, despite Anthropic's safeguards.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.