In a dramatic turn of events that critics say was entirely predictable, Anthropic was forced to disable two of its most advanced AI models hours after a U.S. government export control directive barred foreign nationals from using them. The company, once hailed as the responsible steward of frontier artificial intelligence, now finds itself at the center of a regulatory storm that could reshape how powerful AI systems are developed and deployed.

The affected models—Claude Fable 5 and Mythos 5—belong to Anthropic's "Mythos class," a family of AI systems the company itself described as possessing capabilities so dangerous that they warranted extraordinary precautions. Just weeks ago, Anthropic had warned that these models could aid in developing biological weapons, cyberattacks, and other catastrophic threats. Yet on June 13, 2026, the company announced it had "abruptly disabled" them in response to a U.S. government order, while insisting the decision was based on a misunderstanding.

"We believe this is a misunderstanding and are working to restore access as soon as possible," read the statement on Anthropic's website. The company further claimed that the alleged security vulnerability—a jailbreak technique that bypassed safety guardrails—was "minor" and that other publicly available models could produce the same results without any bypass.

The irony of the situation is not lost on industry observers. Anthropic spent the spring of 2026 building a narrative of existential risk around its Mythos class, releasing detailed system cards and partnering with cybersecurity firms to study the models' potential for harm. The company's Project Glasswing invited limited partners to explore how the models could reshape cybersecurity—almost as if daring the world to pay attention. And pay attention it did.

Mainstream media outlets, from the New York Post to The New York Times, picked up the story. Computer scientist Roman Yampolskiy told the Post that Mythos heralded a new era of AI-powered weapons: "hacking tools, biological weapons, chemical weapons, [and] novel weapons we can't even envision." The phrase "Weapons we can't even envision" became a tabloid headline. British government officials scrambled to form action plans. The Trump administration, previously noninterventionist on AI, signed a safety-focused executive order—reportedly influenced by the mere existence of Mythos.

Then, last week, Anthropic did something seemingly contradictory: it released Claude Fable 5, a "Mythos-class model that we've made safe for general use," according to the company's blog post. Mythos 5 followed, with a limited release as part of Project Glasswing. Brian Merchant, writing in Blood in the Machine, captured the absurdity: "After sparking a major news cycle in the tech media with its April announcement that it had built an AI model so powerful, so dangerous that it threatened to upend the entire civilizational order—and that it was diligently withholding the product from the public so as to protect us from it—the nation's now-1 AI startup decided to put Mythos up for sale after all."

Hours after Merchant's words appeared in print, the export control directive landed at Anthropic's headquarters. The order specifically targeted foreign nationals, prohibiting them from using the models anywhere in the world—including inside the United States. Given that many Anthropic employees themselves are not U.S. nationals, and the global nature of cloud-based AI services, the practical effect was to compel the company to pull both models entirely until the legal situation could be resolved.

Anthropic's statement argued that the government was reacting to a jailbreak technique that the company had already considered and mitigated. "Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities," the statement read. "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."

Yet when Anthropic originally released Fable 5, its own blog post acknowledged that "it is likely impossible to completely prevent universal jailbreaks, but our goal is to make any remaining jailbreaks sufficiently slow and costly that we can detect and prevent them before they are used at scale." The company was aware that absolute safety was unattainable, and it had designed its safety architecture accordingly—including enhanced user data retention and monitoring for abuse. But the government, having been primed for months to fear these models, apparently saw the jailbreak as an unacceptable risk.

Anthropic now warns that government actions like this could, if they became standard, "halt all new model deployments for all frontier model providers." The company's complaint has some merit: the same government that applauded Anthropic's transparency about risks is now punishing the company for the very risks it disclosed. But critics argue that Anthropic brought this upon itself by sensationalizing its own technology to gain attention and regulatory influence, only to treat it as a commercial product once the spotlight faded.

The saga raises deeper questions about the future of AI regulation. How can companies honestly communicate the dangers of powerful AI without inviting overreaction? If every concerning capability must be hidden to avoid government intervention, the transparency that safety advocates have long demanded may become impossible. Conversely, if companies are allowed to hype their models as existential threats and then sell them to the highest bidder, public trust in the entire industry will evaporate.

Anthropic's experience also underscores the challenge of international AI governance. The export control directive applies to foreign nationals, but AI models do not respect borders. A model accessible in the United States can be used by a foreign national sitting in New York just as easily as by one in Beijing—unless the company implements draconian identity verification that could stifle innovation and alienate global talent.

In the immediate aftermath, Anthropic has scrambled to negotiate with U.S. and U.K. authorities, the same governments it had previously worked with to design safeguards. The company hopes to restore access to Fable 5 and Mythos 5 soon, but no timeline has been given. Meanwhile, the incident has sent shockwaves through the AI industry, with other frontier labs reassessing their own communication strategies. Some are reportedly considering more conservative disclosure policies, fearing that the government's reaction to Anthropic may set a precedent.

The story is still unfolding, but one thing is clear: the era of AI companies warning the world about their own creations and then trying to sell them is entering a new, more complicated phase. Whether that phase leads to better safety or simply more secrecy remains to be seen.

Source: Gizmodo News