Trump Shut Down Anthropic's Most Powerful AI Model in 90 Minutes — What the Fable 5 Ban Means for the Tech Industry

Introduction

On the evening of June 12, 2026, Anthropic received a letter. The time stamp on that letter was 5:21 p.m. Eastern Time. By the time most of America sat down for dinner, the company had pulled its two most powerful artificial intelligence models — Fable 5 and Mythos 5 — offline for every single user on earth. The process took roughly ninety minutes. No court order. No public hearing. No advance warning. Just a directive from the Commerce Department, a brief phone call that came in at 1:00 p.m. that same afternoon, and a company scrambling to comply before the close of business.

That is not a hypothetical about the future of AI governance. It already happened. And the questions it raises — about government overreach, about competitive sabotage, about what it truly means to deploy AI commercially in the United States in 2026 — matter to every technology company operating in this country, not just Anthropic.

What Are Fable 5 and Mythos 5?

To understand why this shutdown was such a shock to the industry, you first need to understand what was taken offline.

Anthropic launched Fable 5 and Mythos 5 on June 9, 2026, to significant fanfare. Mythos 5 was Anthropic's most capable AI model ever built, available only to a select group of enterprise partners. Fable 5 was its public-facing counterpart — a version of the Mythos architecture equipped with additional safeguards designed to block responses involving cybersecurity, biology, and chemistry. Anthropic described Fable 5 as a model whose capabilities "exceed those of any model we've ever made generally available." It ranked first on Datacurve's DeepSWE autonomous coding benchmark, scoring 70% PASS@1, three points ahead of OpenAI's GPT-5.5. In drug discovery trials, Mythos 5 was reported to have accelerated research timelines by a factor of ten. It matched or outperformed experienced scientists on protein design tasks.

These were not incremental upgrades. They represented what Anthropic called a new tier of capability, which it labelled "Mythos-class" — a level above its previous Opus-class models. The models were the product of extensive pre-release testing, including thousands of hours of red-teaming with the US government, the UK AI Safety Institute, and multiple private third-party organizations. Anthropic had worked directly with government agencies before the launch and, according to a person familiar with those discussions, received approval to deploy both models.

The Jailbreak That Wasn't — And the Amazon Angle

The official trigger for the shutdown was a jailbreak. According to Anthropic, the government told the company it had become aware of a method of bypassing Fable 5's cybersecurity guardrails. The specific technique, as described by cybersecurity expert Katie Moussouris of Luta Security — who reviewed the underlying research paper — amounted to asking the model to read a specific codebase and fix any software flaws, after it initially refused to review the code for security issues directly.

Anthropic pushed back hard on the characterization. The company described the vulnerability as a "potential narrow, non-universal jailbreak" that did not unlock broad dangerous capabilities. It said the same technique could be used to draw similar responses from other publicly available models, including OpenAI's GPT-5.5 and even Chinese models. No tester had found a universal jailbreak that could broadly bypass Fable 5's safeguards. The company had only received verbal notice of the concern, with no specific technical details disclosed in the written directive itself.

The entity that flagged the vulnerability to the government was not a neutral researcher. Multiple reports confirmed the paper was authored by security researchers at Amazon. Amazon's CEO Andy Jassy personally escalated the findings to Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross. That is a striking chain of events, because Amazon is also Anthropic's largest investor, having committed up to $25 billion in the company. Amazon is Anthropic's primary cloud host. Amazon also relies on Anthropic's models for many of its own services. The fact that a competitor and investor simultaneously flagged a security concern through the highest levels of the executive branch has led multiple observers to question whether commercial rivalry played a role in the government's response, even if unintentionally.

A Feud That Started Long Before June 12

The export control directive did not arrive in a vacuum. The relationship between Anthropic and the Trump administration had been deteriorating for months before Fable 5 ever launched.

In early 2026, the Pentagon approached Anthropic with a contract that required the company to agree that its AI models could be used for any "lawful purpose." Anthropic refused, seeking specific exemptions to ensure that Claude could not be used for fully autonomous weapons systems or mass domestic surveillance. Those were not vague philosophical concerns. They were narrow contractual carve-outs on two specific use cases. Negotiations broke down.

In late February 2026, Defense Secretary Pete Hegseth announced on X that Anthropic was being designated a "Supply-Chain Risk to National Security." The post declared that no contractor, supplier, or partner doing business with the United States military could conduct any commercial activity with Anthropic. President Trump followed shortly after with a Truth Social post ordering all federal agencies to immediately stop using Anthropic's technology. The irony, noted by many observers, was that the DOD supply chain risk designation — a label historically reserved for foreign adversaries like Huawei — was being applied to an American company for the first time in history.

The Pentagon formally issued the designation in early March. Anthropic filed suit in federal court, challenging the designation as unconstitutional, arbitrary, and retaliatory. An appeals court denied the company's request for a temporary stay. The legal battle was still ongoing when Fable 5 launched.

What Hegseth Said, and What Sacks Claimed

After the June 12 shutdown, the administration's messaging was fragmented and, at points, contradictory. Commerce Secretary Howard Lutnick sent the written directive to Anthropic CEO Dario Amodei, citing the Export Control Reform Act of 2018 and stating the department was acting over concerns that the models could be accessed by military and intelligence users in China, Russia, or other countries of concern. Lutnick warned that failure to comply would result in criminal and civil penalties.

David Sacks, co-chair of the President's Council of Advisers on Science and Technology, offered a different account on X. He claimed the administration had given Anthropic a choice before issuing the formal directive: fix the jailbreak or voluntarily take the model down. According to Sacks, Amodei declined both options. "The admin issued this reluctantly," Sacks wrote, adding that the administration had been "very surprised that Anthropic hasn't wanted to cooperate with a reasonable safety request." Anthropic disputed that characterization, maintaining that the vulnerability was too narrow and too widely replicable across other models to justify pulling a product deployed to hundreds of millions of people.

An additional wrinkle emerged when Axios reported that the White House's frustration was not limited to the jailbreak itself. Officials apparently took issue with the fact that Anthropic had engaged Moussouris to independently review the Amazon findings — and the administration reportedly viewed Moussouris as a "radical Democrat," a characterization that raised serious questions about whether scientific credibility or political alignment was driving the government's technical judgments.

The Cybersecurity Community Responds

Within 48 hours, the backlash from the security research community was significant. A group of leading cybersecurity professionals published an open letter organized by Alex Stamos, former chief security officer at Facebook and now chief product officer at Corridor. The letter called on Commerce Secretary Lutnick and National Cyber Director Cairncross to lift the export controls immediately. It was ultimately signed by roughly 100 to 150 cybersecurity professionals from companies including Nvidia, Adobe, Zoom, Google, Sophos, and Anaplan, as well as several academic researchers.

The letter's core argument was direct: taking the most capable AI models away from defenders does not stop adversaries — it stops defenders. Security professionals use frontier AI models every day to hunt for vulnerabilities before attackers find them, to write detection rules, and to analyze malware at speed. Adversaries, the letter noted, could use open-source alternatives, foreign models, or older techniques. The US government had achieved nothing except handicapping the very people responsible for protecting American software infrastructure. "To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous," the letter stated plainly.

The timing and consequences of the ban gave that argument immediate force. The day after Fable 5 was taken down, Chinese AI company Zhipu AI launched its GLM-5.2 model and explicitly cited the US ban as evidence that American AI models could not be relied upon. Zhipu's stock surged 33% on the announcement.

Negotiations, the G7, and a Thaw That Has Not Yet Produced a Restoration

In the days following the shutdown, Anthropic's senior technical staff met with officials at the Commerce Department in Washington virtually every day. Commerce Secretary Lutnick held regular calls with Anthropic leadership. Dario Amodei attended the G7 summit in Évian-les-Bains, France, where he met directly with President Trump. At a working lunch for G7 leaders and tech CEOs on June 17, Amodei — alongside Google DeepMind CEO Demis Hassabis — proposed a US-led democratic AI alliance to coordinate standards and exclude China from frontier AI access.

Trump, speaking from the summit's sidelines, said negotiations with Anthropic were "going fine" and described Amodei as "nice" and "smart." He added that he no longer viewed Anthropic as a national security threat, though he noted that a week earlier, he might have. Separately, the White House and Anthropic were reported to be drafting a joint risk framework to assess the severity of future AI security flaws and guide when government intervention would be appropriate. Anthropic's international managing director stated publicly that he was "very confident" both models would return "in the coming days."

As of this writing, Fable 5 and Mythos 5 remain offline. The directive has not been formally withdrawn.

What This Means for Every AI Company in the United States

The most important question raised by the Fable 5 shutdown is not specific to Anthropic. It is this: under what rules does any AI company in the United States now operate?

The Commerce Department used the Export Control Reform Act of 2018 to issue its directive — a law originally designed to govern physical technology exports. Export control experts noted almost immediately that AI models are not technically "exported" in the traditional sense; they are accessed remotely. Whether the government had legal authority to apply these controls to cloud-based AI access remains legally contested. Yet the directive landed, the company complied, and the models went dark worldwide before a single court had reviewed the action.

The precedent is stark. A government can, using existing statutory authority, order a technology company to disable its flagship product within ninety minutes, for every user on earth, without a detailed public justification, without prior notice, and without any transparent process. Anthropic itself had, just two days before the shutdown, published a policy essay by Amodei calling for the government to have legal authority to block unsafe AI deployments. Two days later, the government used exactly that type of authority against the company — not through the transparent, technically grounded process Amodei had envisioned, but through an emergency directive that Anthropic described as lacking specific details and not adhering to any established principles.

The consequences extend beyond one company's regulatory dispute. French President Macron warned publicly that no country would purchase American AI if it could be switched off at any moment. Canadian Prime Minister Mark Carney said the incident was a lesson in the danger of over-reliance on a single nation's technology. UK AI minister Kanishka Narayan called it evidence that Britain needs greater sovereign AI capability. The ban accelerated international calls for countries to build their own AI infrastructure, which is precisely the opposite of what a US AI competitiveness strategy should produce.

For every AI company currently building a frontier model or deploying one commercially in the United States, the Fable 5 shutdown establishes several uncomfortable facts. A commercial model with hundreds of millions of users can be pulled offline unilaterally. A competitor with financial ties to a target company can trigger a national security review. A narrow jailbreak that other models share equally can be used to justify singling out one provider. And the process for challenging such a directive — litigation, lobbying, international diplomacy — is slow, expensive, and provides no guarantee of rapid restoration.

Conclusion

The Fable 5 ban is not primarily a story about a jailbreak. It is a story about power — specifically, who holds it, who does not, and how quickly it can be exercised. A company valued at close to a trillion dollars, with months of pre-launch government coordination and thousands of hours of safety testing behind it, was taken offline in ninety minutes by a letter that did not explain its specific concern and was not subject to immediate judicial review.

The deeper lesson for the industry is that technical compliance with safety standards does not protect a company from political consequences. Anthropic built genuine safeguards, published transparent safety documentation, cooperated with multiple government agencies before launch, and still found itself shut down. The variables that ultimately mattered were not technical. They were a long-running political dispute, a competitor's escalation to the highest levels of the executive branch, and a government that demonstrated it is willing to act unilaterally at speed.

Every AI company operating in the United States should treat the Fable 5 shutdown as a stress test they did not know they were taking. The question they now need to answer is not whether this could happen to them. The question is whether they have a plan for the 90 minutes after it does.