Anthropic’s Fable 5 and Mythos 5 remain unavailable due to a Trump administration order, and the situation has escalated rather than subsided. It’s practically a Hollywood movie plot at this point. The Trump administration is now demanding that Anthropic relaunch Fable 5 with safeguards that cannot be circumvented by jailbreaks. However, AI experts note that no artificial intelligence model can offer absolute guarantees, as such vulnerabilities are constantly evolving and cannot be fully controlled. This effectively means that until a more robust solution is developed and approved, Fable 5 may remain inaccessible globally.
The debate surrounding this issue spans multiple angles: a government publicly pressuring a private company that refused direct cooperation, restricting paying users’ access to more powerful AI with an unresolvable excuse. Is it dystopian? Apparently, not dystopian enough.
Trump Threatens Anthropic Over Fable 5 and Mythos 5: They Must Guarantee No Jailbreaks… Which is Impossible
The confrontation occurred just days after the public launch. As previously reported, Anthropic introduced Fable 5 as a general-purpose version based on Mythos, while Mythos 5 was reserved for cybersecurity and critical infrastructure defenders within Project Glasswing.
Anthropic received the directive on June 12th at 5:21 PM ET. The letter did not specify a precise national security reason, though Anthropic inferred that the government believed it had found a method to bypass Fable 5’s protections. The company asserts that it reviewed a demonstration used to identify a few minor, already known vulnerabilities.
The shutdown affected Fable 5 and Mythos 5 for all customers, as the order mandated cutting off access to any foreign national, even within Anthropic itself. Other Claude models were unaffected, meaning the impact was concentrated on the two most sensitive offerings in the new catalog.
Selective Access for Businesses
The situation is now a point of negotiation, with discussions reportedly focused on a common framework between the White House and Anthropic to measure the severity of security flaws. This system would consider the bypassed safeguards, the exposed capabilities, and the practical consequences of each breach.
The technical nuances are significant because a jailbreak isn’t a permanent off switch; it’s an attack surface that shifts with prompts, models, filters, and automated testing. Therefore, blocking all jailbreaks is fundamentally incompatible with how these systems operate, as it’s virtually impossible given the near-infinite possibilities. So much so that, according to WIRED, even the NSA concluded that there are ways to disable these protections, and there always will be.
Reuters added another important detail last night: some companies that tested Mythos Preview within Glasswing, such as Dragos and Cisco Systems, reportedly retained access, while ENISA was informed it would not receive access for now. Anthropic has not publicly clarified its case-by-case decision-making process.
It remains to be seen what Washington will accept as sufficient proof. If they demand total security, Fable 5 and Mythos 5 could remain blocked for a considerable time, likely until Trump is presented with a demonstrably superior model capable of detecting and correcting vulnerabilities far beyond these two AIs.
If they accept technical metrics, Anthropic may find a way to relaunch without promising something that is currently impossible to guarantee. However, the reality is that Fable 5 and Mythos 5 are as blocked by Trump as the Strait of Hormuz. We are witnessing politics, once again, obstructing private companies, but this time without government collaboration, which makes the situation worse. In the West, we criticize China, but we are increasingly resembling them, ironically.
