{"id":56767,"date":"2026-06-15T20:56:49","date_gmt":"2026-06-16T00:56:49","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=56767"},"modified":"2026-06-15T20:56:49","modified_gmt":"2026-06-16T00:56:49","slug":"us-government-unhackable-llms-anthropic","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/us-government-unhackable-llms-anthropic\/","title":{"rendered":"US Government Demands Unhackable LLMs from Anthropic"},"content":{"rendered":"

The Trump administration<\/a> is demanding that frontier AI models be effectively unhackable before they can be released, placing Anthropic squarely in the crosshairs after the company shipped Fable 5<\/a> without waiting for a government clearinghouse that has yet to be stood up. The confrontation, which has drawn in the CIA, the Department of Commerce, and over 100 cybersecurity experts, reveals a fundamental disconnect between how the government believes large language models should be secured and how the industry knows they actually work.<\/p>\n

The core dispute centers on the administration’s cyber executive order, which called for voluntary government oversight of AI models. Anthropic publicly welcomed the proposal but proceeded to release Fable 5 before the designated review body could be established. A government official characterized the move bluntly: “They came to every fork in the road and took the wrong fork.” The accusation that Anthropic knew about a potential jailbreak and stayed silent has since become the administration’s principal complaint, though the existence and severity of that specific vulnerability have not been confirmed.<\/p>\n

Why the Jailbreak Accusation Reveals More About the Government Than About Anthropic<\/h2>\n

The claim that Anthropic should have known a jailbreak was possible and withheld the model says far more about the government’s understanding of large language model security than about Anthropic’s conduct. Anyone who works closely with frontier AI systems understands that prompt injection and jailbreaking are not bugs waiting to be fixed; they are inherent properties of the technology. OpenAI has stated openly that prompt injection may never be fully solved. There is no known method for making an LLM truly unhackable, and no company\u2014American or otherwise\u2014has shipped a model that meets that standard.<\/p>\n

The real question is not whether a jailbreak exists but how severe the exploit is and how quickly countermeasures can be deployed. The administration’s insistence on an “unhackable” standard before any model can ship internationally sets the stage for difficult negotiations, not just with Anthropic but with every company building at the frontier. Anthropic’s own CEO, Dario Amodei, acknowledged as far back as 2023 that a jailbreak bypassing safety protocols in science, technology, or biology could become a matter of life and death. That admission now cuts both ways: it validates the need for robust safeguards, but it also underscores that no model is immune.<\/p>\n

Cybersecurity Community Rallies Behind Anthropic With an Open Letter<\/h2>\n

More than 100 security experts and technology executives have published an open letter to Trade Secretary Lutnick and National Cyber Director Cairncross, calling for the immediate lifting of export controls on both Fable and Mythos<\/a>. The signatories argue that while Anthropic’s models are effective at finding software vulnerabilities, they are not uniquely capable. Other models\u2014including GPT-5.5, Opus, Sonnet, and the Chinese model Kimi 2.7\u2014can perform the same security analysis work.<\/p>\n

The letter takes direct aim at the logic underpinning the export restrictions. Anthropic built several safety mechanisms into Fable that the security community largely dismissed as overkill at launch. The signatories warn that export controls risk stripping American defenders of the best available tools while Chinese open-weight models are only months behind the top U.S. systems. Among those who signed are Alex Stamos<\/a> of Corridor, Rachel Tobac of SocialProof Security, Katie Moussouris of Luta Security, Dan Lorenc of Chainguard, and Joe Levy of Sophos.<\/p>\n

A Negotiation Conducted in Different Languages<\/h2>\n

Government sources have described the communication between the two sides as fundamentally mismatched. “It’s like they just speak in different languages,” one official told Axios<\/a>. The Department of Commerce and Anthropic employees are currently in talks, with additional meetings planned that will involve the CIA and science advisor Michael Kratsios. The appointment of a director for the AI Safety Institute’s clearinghouse remains pending, leaving the oversight body the executive order envisioned still not operational.<\/p>\n

What This Means for Developers and Security Teams<\/h2>\n

For professionals building on or evaluating frontier AI models, the immediate takeaway is that no model should be treated as impervious to adversarial input. The government’s demand for an unhackable standard is not aligned with the current state of the technology, and organisations should plan their own security postures accordingly. Whether or not Fable 5 contained a specific vulnerability that warranted delay, the broader pattern is clear: every frontier model carries inherent jailbreak risk, and the responsible approach is to deploy with monitoring, rapid response, and realistic expectations rather than waiting for a standard that does not yet exist. Those evaluating AI tools for security-critical workflows should prioritise models with transparent incident response mechanisms and a demonstrated willingness to patch quickly, regardless of which company built them.<\/p>\n","protected":false},"excerpt":{"rendered":"

The Trump administration is demanding that frontier AI models be effectively unhackable before they can be released, placing Anthropic squarely in the crosshairs after the company shipped Fable 5 without waiting for a government clearinghouse that has yet to be stood up. The confrontation, which has drawn in the CIA, the Department of Commerce, and […]<\/p>\n","protected":false},"author":7,"featured_media":84781,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/56767.png","fifu_image_alt":"US Government Demands Unhackable LLMs from Anthropic","footnotes":""},"categories":[349],"tags":[],"class_list":["post-56767","post","type-post","status-publish","format-standard","has-post-thumbnail","category-articles"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/56767.png","fifu_image_alt":"US Government Demands Unhackable LLMs from Anthropic","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/56767","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=56767"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/56767\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/84781"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=56767"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=56767"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=56767"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}