US Disables Anthropic’s Mythos and Fable

A US directive citing security concerns forced Anthropic to disable its recently released Fable 5 and Mythos 5 AI models, worrying many working in cyber. 

The latest chapter in the saga between Anthropic and the US government unfolded last week, sparking even more conversations about the role of artificial intelligence (AI) in government cybersecurity and defense. On June 12, 2026, Anthropic released a statement explaining that on 5:21 pm (ET) that Friday, they had received a directive from the White House “to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.” In the statement, Anthropic shared that the order essentially forced them to “abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.”

Those following the ongoing saga between AI giant Anthropic and the US may already be familiar with Mythos. Mythos AI made headlines in April as a cybersecurity tool that was too powerful for commercial release. The model reportedly uncovered “thousands of high-severity vulnerabilities, including some in every major operating system and web browser” - 99% of which had not yet been patched. This ability presented a clear risk if the tool fell into the wrong hands, leading to Athropic stopping a commercial release of the tool and instead opting for a smaller release with their Project Glasswing partners, like Amazon, NVIDIA, Google, and Microsoft. Though an investigation into access to the tool by 3 unauthorized users stoked some concerns about the tool being used by threat actors, Mythos AI was quickly integrated by the Glasswing partners and federal agencies in the US and UK. Expansion has been accelerating as of late, with Anthropic announcing the addition of 100 new organizations (on top of the roughly 50 groups at the initial rollout), largely involved with critical infrastructure, across 15 countries, in early June.

This expansion was at odds with the tense relationship between the US government and Anthropic. In February, the company’s refusal to roll back some of their safety standards in their contracted work with the administration - standards that would prevent AI-directed warfare and domestic mass surveillance - put them in a public feud with Pete Hegseth and the Department of War. They were essentially blacklisted from government-related work for their position, and uptake of Mythos AI in federal spaces has been a notable exception to the policy change - until now.

Fable 5 is one of the latest tools from Anthropic AI, released on June 9th, 2026, alongside Mythos 5. In the press release, Anthropic touted the Fable 5 as “a Mythos-class model that we’ve made safer for general use.” The key difference between the models is that Fable 5 includes conservatively tuned safeguards so that certain prompts will trigger responses from Claude Opus 4.8 (an older AI model), and Mythos 5 is essentially the same model as Fable 5, but with fewer safeguards in some areas. According to the announcement, Mythos 5 “had the strongest cybersecurity capabilities of any model in the world” and was set to be released in collaboration with the US government.

With the new directive, the collaboration has been put into question. According to Anthropic, the directive came from a belief that the guardrails of Fable 5 could be bypassed through manipulation, making it a potential cybersecurity risk. For Anthropic, this concern about “jailbreaking” one of their AI models is not a new one. In February, news broke about Anthropic’s chatbot, Claude, being manipulated into giving a hacker access to 150 GB of sensitive data from Mexico’s government. The hacker overrode the bot’s guardrails and gained “thousands of detailed reports that included ready-to-execute plans, telling the human operator exactly which internal targets to attack next and what credentials to use”. With this incident in mind, the directive may not sound surprising. Still, the public reaction to Fable 5 did not seem to completely align with the directive’s characterization of the risk posed by the model.

Following the release of Fable 5, several cybersecurity professionals claimed the model’s guardrails were actually too conservative to make it useful for those in the industry. Valentina “Chompie” Palmiotti, the head of X-Force Offensive Research at IBM, said on X that the model “rejects any request that could be tangentially cyber related. Even innocuous tasks like reading a blog post.” Another cybersecurity veteran noted the model “seems to be keyword based, so anything in the lexical field of ‘cybersecurity’ triggers the guardrails.”

The reports that Fable 5 was still jailbroken seemed not to necessarily hinge on a failure of word-based guardrails, but rather simply asking the model to fix code and script tests for patches on a code with built-in vulnerabilities. One cybersecurity expert, Katie Moussouris, the only outsider who saw the government report on the Fable 5 bypasses, did not qualify the method as a jailbreak: “Defenders need to be able to ask AI to fix the bugs in a file, explain why the fix matters, and write tests that confirm the patch works. That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security: executing the find, fix, and test loop defenders run every day.”

Since the June 12th decision from the government, more have come out echoing Moussouris’ feelings on the ban that she called “misguided” and “heavy-handed and hasty”. Several cyber leaders have banded together to oppose the directive, with over 160 CEOs, researchers and security officers signing a letter, addressing the new risks that came with the ban: “The Chinese open-weight models are only months behind the best American models, and those are the models we know about. It seems likely that the PRC government has access to private capabilities beyond what has been published. To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous.”

Anthropic has made its stance on the directive clear, saying “we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.” They aren’t the only ones who have found the government’s actions suspicious, considering the fallout after Anthropic’s fight with Hegseth. Democratic Senator Mark Warner spoke with Cyberscoop about the situation: “This administration has repeatedly shown a willingness to weaken export controls designed to protect our national security and maintain our technological edge over adversaries, while also making no secret of its hostility toward Anthropic. That raises serious questions about whether this effort is being driven by objective national security concerns or something else.”

Whether politically-motivated or security-guided, the directive will be a milestone for the US as it continues to compete in the global AI arms race.