- Anthropic Fable 5 and Mythos 5 have been blocked for all customers following a US government national security directive issued June 12.
- The Anthropic Fable 5 suspension covers all foreign nationals, including Anthropic’s own employees, regardless of where they are located.
- The government cited intelligence about a potential narrow jailbreak method targeting the newly launched cybersecurity-capable model.
- Anthropic openly disagrees with the government’s approach, arguing the action lacks transparency and proper technical grounding.
- Anthropic Fable 5 and Mythos 5 have been blocked for all customers following a US government national security directive issued June 12.
- The Anthropic Fable 5 suspension covers all foreign nationals, including Anthropic’s own employees, regardless of where they are located.
- The government cited intelligence about a potential narrow jailbreak method targeting the newly launched cybersecurity-capable model.
- Anthropic openly disagrees with the government’s approach, arguing the action lacks transparency and proper technical grounding.
Table of Contents
Anthropic Fable 5 Goes Dark Overnight
Anthropic Fable 5, the company’s most capable AI model to date, was cut off from all customers late last week after the US government issued a directive on Friday, June 12, ordering Anthropic to suspend access on national security grounds. The block extends to Mythos 5 as well — the state-of-the-art cybersecurity model that underpins Fable’s capabilities. Every other Anthropic product, including its widely used Claude chatbot, remains online and unaffected.
The timing is striking. Anthropic had only launched Fable publicly on June 9 — just three days before the government’s order arrived. That’s barely enough time for most enterprise customers to finish onboarding, let alone for regulators to conduct any kind of deep technical review. Whatever intelligence triggered this directive moved fast.
What Is Fable, and Why Does It Matter?
To understand why this story carries real weight, you need a bit of background on what Anthropic was actually shipping. Mythos is Anthropic’s restricted, high-capability cybersecurity model — it’s not available to the general public and exists only within the company’s Project Glasswing partner ecosystem. Think of it as the classified tier of Anthropic’s model lineup, built for sophisticated security use cases that Anthropic apparently didn’t want running in the open.
Fable was designed to bring a meaningful slice of Mythos’s power to a much wider audience. According to Anthropic, Fable’s capabilities “exceed” any previous model the company has released — and it offered a telling benchmark to illustrate that point. During internal testing, Anthropic Fable 5 successfully completed Pokémon FireRed. Claude, by contrast, failed to beat Pokémon Red, the game that FireRed was based on. It’s a quirky benchmark, but the underlying message is clear: Anthropic Fable 5 represents a genuine capability step-change, particularly in long-horizon reasoning and autonomous task completion.
That’s precisely what makes cybersecurity researchers — and apparently, government officials — pay close attention. A model that can autonomously solve complex, multi-step problems is enormously useful for defensive security work. It’s also, in the wrong hands, exactly the kind of tool that could be pointed at offensive targets. Anthropic’s own AI safety research has consistently acknowledged this dual-use tension as one of the central challenges in deploying frontier models responsibly.
The Jailbreak Concern at the Centre of This
Anthropic says the government hasn’t formally spelled out its national security rationale, but the company has a working theory: officials received information about a method for jailbreaking Anthropic Fable 5. Specifically, Anthropic describes it as a potential narrow, non-universal jailbreak — meaning it would unlock restricted behaviours in Fable but wouldn’t automatically transfer to other models or represent a systemic vulnerability across Anthropic’s entire product range.
Anthropic was candid in its public statement about the inherent limits of any AI safety regime.
“We aimed to make jailbreaks either narrow (in the case of non-universal jailbreaks) or very expensive to produce (in the case of universal jailbreaks), and to combine this with thorough monitoring to quickly detect and shut down any successful attacks.”That’s an honest framing, and it reflects something the AI industry has been grappling with for years: no model is permanently jailbreak-proof. The question is always whether the cost and effort required to break the safeguards outpaces the attacker’s resources and motivation.
The company also noted, with some frustration implicit in the phrasing, that its safeguards for Anthropic Fable 5 were aggressive enough that “many users have complained that they are overly broad.” In other words, Anthropic had already erred on the side of restriction — only for a government order to pull the entire product anyway.
Who Gets Cut Off — and the Scope Is Broader Than It Sounds
The government’s directive, as Anthropic describes it, was targeted specifically at foreign nationals. But the practical implementation is sweeping: Anthropic has suspended access for all foreign nationals, whether they’re located inside or outside the United States, and — notably — even if they’re Anthropic employees. That last detail is significant. It means Anthropic may have had to lock some of its own staff out of models they were actively working on, pending whatever resolution comes next.
For Anthropic’s enterprise customers, particularly those operating globally with international teams, the implications are immediate and disruptive. A directive framed around ‘foreign nationals’ can effectively hobble a multinational company’s AI workflows overnight, with no clear timeline for resolution.
Anthropic Pushes Back — and That’s the More Important Story
Here’s where this gets genuinely interesting from an industry perspective. Anthropic complied with the order. But the company did not stay quiet about it, and its public statement reads as a pointed rebuke of how the government handled this intervention.
Anthropic has long been among the more vocal AI labs on the question of government oversight — it has consistently argued that the industry needs clearer regulatory frameworks and that AI companies shouldn’t be left to self-police indefinitely. In principle, the company supports the government having the ability to block unsafe AI deployments. But it drew a clear line in its statement:
Source: Engadget
