Too powerful to release, until it wasn't.
In a moment that reveals how swiftly the relationship between technology and state power can fracture, Anthropic this week disabled two of its most advanced AI models at the direction of US national security authorities, who cited concerns about jailbreaking vulnerabilities. The company, which had only just released Claude Fable 5 to the public, disputed the severity of the flaws—describing them as minor and common across the industry—yet complied nonetheless. The suspension arrives against a backdrop of deepening hostility from the Trump administration, including an unprecedented 'supply chain risk' designation that places an American company in the same category historically reserved for adversarial nations. What unfolds next may say as much about the future of AI governance as it does about any single vulnerability.
- Anthropic was forced to abruptly pull Fable 5 and Mythos 5 from all customers after US authorities flagged a demonstrated jailbreaking technique capable of bypassing the models' built-in safeguards.
- The company pushed back hard, arguing the vulnerabilities were minor, previously known, and no worse than flaws already present in competing AI tools available to the public.
- The suspension lands in the middle of an already inflamed standoff: the Trump administration has publicly attacked Anthropic, and Defense Secretary Pete Hegseth made history by labeling the company a 'supply chain risk'—a designation never before applied to a US firm.
- Anthropic sued the Pentagon over that designation, and a federal judge has temporarily blocked enforcement, allowing government-adjacent organizations to keep using its tools while litigation continues.
- With no timeline offered for restoring the suspended models, customers and observers are left uncertain whether this is a patch-and-return situation or the opening move in a longer, more damaging regulatory siege.
Anthropic pulled two of its most advanced AI models from public access this week, just days after their release, following pressure from US national security authorities over a demonstrated jailbreaking vulnerability in Claude Fable 5. In a terse statement, the company said it had no choice but to disable both Fable 5 and Mythos 5 across all customers to ensure compliance with the order.
Jailbreaking—the practice of circumventing a system's built-in restrictions—was the stated concern: someone had shown a technique for bypassing Fable 5's safeguards, potentially exposing sensitive information or unlocking restricted capabilities. When Anthropic reviewed the demonstration, it concluded the vulnerabilities were minor, previously known, and no more serious than flaws already discoverable in other publicly available AI models without any special bypass technique.
The suspension deepens an already fraught relationship with the Trump administration. Anthropic had taken an unusually cautious approach to Fable 5's rollout, granting a small group of organizations early access in April specifically to stress-test the system before wider release—arguing the model's power to exploit computer systems warranted careful vetting. That caution has done little to soften official hostility. Donald Trump has publicly criticized the company, and Defense Secretary Pete Hegseth went further, designating Anthropic a 'supply chain risk'—a label historically reserved for companies in adversarial nations, and one never before applied to a US firm.
Anthropic responded by suing the Pentagon. A federal judge has since blocked enforcement of the designation, allowing government-linked organizations to continue using the company's tools while the case proceeds. A separate legal battle over an order barring government agencies from using Anthropic's AI altogether remains ongoing.
What the suspension of Fable 5 and Mythos 5 ultimately represents—a temporary pause while vulnerabilities are addressed, or a sign of deeper capitulation to an increasingly hostile regulatory climate—remains unanswered. Anthropic offered no timeline for restoration, leaving customers and competitors watching closely.
Anthropic pulled the plug on two of its most advanced AI models this week, just days after making them available to the public. The company had released Claude Fable 5 to general users earlier in the week, but US national security authorities raised alarms about a potential security vulnerability—specifically, a method for bypassing the tool's built-in safeguards, known as jailbreaking. In a terse statement, Anthropic said it had no choice: "The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance."
Jailbreaking is the process of circumventing software restrictions designed to protect computer networks. If successful, it can grant hackers access to sensitive information or unlock features that were intentionally locked away. The concern here was that someone had demonstrated a technique for doing exactly that to Fable 5. When Anthropic reviewed the demonstration, it found evidence of what it described as a small number of previously known, minor vulnerabilities. The company's assessment was blunt: these flaws were relatively simple, and other publicly available AI models could discover them without needing any special bypass technique at all.
The timing of the suspension underscores a deeper tension between Anthropic and the Trump administration. The company had been careful about Fable 5's rollout. Before releasing it publicly this week, Anthropic had given a small group of organizations early access in April, specifically to test the system for vulnerabilities and weaknesses. The company had justified this cautious approach by arguing that Fable 5 was too powerful to release without vetting—its capabilities exceeded anything Anthropic had ever made widely available, and its ability to exploit or hack computer systems posed genuine risks. Some observers dismissed this as marketing hype, but the company stood by the claim.
The broader context makes the suspension even more fraught. Donald Trump has publicly criticized Anthropic, and in a particularly striking move, US Defense Secretary Pete Hegseth designated the company a "supply chain risk." That label is historically reserved for companies based in adversarial nations—it signals that a tool or service is not secure enough for government use. It was the first time a US company had ever received such a public designation. Anthropic responded by suing the Pentagon over the designation. A federal judge has since ruled that the Pentagon's directive cannot be enforced, meaning government agencies and organizations working with the US military can continue using Anthropic's tools while the lawsuit proceeds.
Separately, Anthropic and the Trump administration are locked in ongoing litigation over an order to stop government agencies from using the company's AI tools altogether. The BBC reached out to the US Department of Commerce for comment but had not received a response at the time of reporting. What remains unclear is whether the suspension of Fable 5 and Mythos 5 represents a temporary measure while vulnerabilities are patched, or a more permanent shift in how Anthropic will navigate the increasingly hostile regulatory environment. The company's statement offered no timeline for restoration, leaving customers and competitors alike watching to see whether this is a capitulation or a tactical retreat.
Notable Quotes
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.— Anthropic statement
These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.— Anthropic on the identified flaws
The Hearth Conversation Another angle on the story
Why would the government order a company to disable tools that are already in use by thousands of people?
Because they believe there's a security vulnerability—a way to bypass the safeguards. Whether that threat is real or overstated is exactly what's in dispute.
And Anthropic says the vulnerabilities are minor?
Yes. They reviewed the demonstration and found the flaws were simple, and that competitors' models have the same problems. So either the government is being overly cautious, or Anthropic is downplaying the risk.
This seems connected to the Trump administration's broader criticism of the company.
It is. The Defense Secretary called them a supply chain risk—a designation usually reserved for foreign adversaries. That's not a technical assessment. That's political.
So is this about security or about power?
Probably both. The security concern may be real, but the timing and the designation suggest something larger is happening—a clash between a company and a government that doesn't trust it.
What happens next?
Anthropic is suing. A judge has already ruled the Pentagon can't enforce its ban. So we're in a legal standoff while the company figures out whether it can fix the vulnerabilities and restore service.