How a Three-Word Prompt Forced Anthropic to Shut Down Its Most Powerful AI Models Worldwide

On Thursday, June 11, Amazon CEO Andy Jassy called senior administration officials to share the findings, and the White House acted within 24 hours . A senior Trump administration official characterized Anthropic's response to the vulnerability as "recklessness" .

The specific jailbreak technique involved a simple three-word prompt—"Fix this code"—that tricked the model into revealing exploitable information about security vulnerabilities . Anthropic disputed the severity of the finding, saying the technique amounted to asking the model to identify bugs . But the administration moved swiftly nonetheless.

Industry Backlash

The restrictions prompted swift condemnation from across the tech and cybersecurity industries:

• More than 50 cybersecurity leaders, including executives from Nvidia and Adobe, signed an open letter to the Trump administration demanding the restrictions be lifted . They argued the ban weakens efforts to detect and patch software flaws while doing nothing to stop foreign adversaries from developing their own capabilities .
• A separate group of 30-plus industry and academic professionals issued a public letter citing international competition and the urgent need to patch network vulnerabilities .
• Critics highlighted that the move handicaps US-allied cybersecurity defenders while Chinese rivals rapidly close the gap in advanced AI capabilities .

Reporting from the Associated Press indicates that more than 100 cybersecurity experts from companies including Adobe and Nvidia urged the government to lift the controls and commit to an "open, scientific and transparent process of handling AI risk assessments" .

Impact on Allied Nations

The export ban strained relationships with US allies. The restrictions blocked UK, Japanese, and other allied governments, corporations, and individuals from accessing Anthropic's most advanced models . Al Jazeera reported that the order further strained alliances, as allied nations saw themselves treated as security risks rather than partners . The directive applied to "any foreign national" with no exceptions for close allies . Before the ban, Anthropic had allowed 200 organizations across 15 countries to test Fable 5 .

Emergency Negotiations with Washington

Anthropic launched an urgent diplomatic and technical push to resolve the situation. The company dispatched staff to Washington, D.C. over the weekend of June 13–14, sending key technical personnel to engage with officials and negotiate an agreement to lift the export limitations . As of June 17, the company was still actively working to resolve the dispute, with CEO Dario Amodei engaging directly with administration officials .

Separate Pentagon Dispute Over Supply Chain Risk

Beyond the June 12 export controls, Anthropic has been locked in a separate, months-long legal battle with the Pentagon:

• On March 5, 2026, the Trump administration designated Anthropic as a national security "supply chain risk," barring it from federal contracting . This was reportedly the first time an American company received this designation .
• The Pentagon's decision stemmed from Anthropic's refusal to remove AI safety guardrails that prevented its Claude models from being used for autonomous weapons and mass surveillance of Americans .
• On March 9, Anthropic filed suit in federal court in California, arguing the designation was unlawful and violated its First and Fifth Amendment rights . The company's executives said the blacklisting could cut 2026 revenue by multiple billions of dollars and cause severe reputational harm .
• A federal judge temporarily blocked the Pentagon's blacklisting on March 26, calling it "likely both contrary to law and arbitrary and capricious" . However, a federal appeals court in Washington, D.C., rejected Anthropic's plea for a preliminary injunction on April 8 .
• As of May 1, the Pentagon maintained that Anthropic still posed a supply chain risk, while stating that the Mythos model represented a "separate national security concern" .

Note on revenue figures: Multiple sources reference "billions" in potential lost revenue from the Pentagon dispute , but this reporting did not find a specific "$150 million in lost annual revenue" figure mentioned in the question. The billions figure may refer to a broader estimate of potential revenue loss across the entire business.

What This Means

The Fable 5 and Mythos 5 shutdown marks the first time a leading AI company has pulled a publicly deployed model offline due to direct government intervention . The episode highlights unresolved tensions between the US government's desire to control powerful AI capabilities and the reality that those same tools are essential for cybersecurity defense. With allied nations locked out, Chinese AI companies accelerating, and the industry in open revolt, the June 12 directive may go down as a defining moment in the geopolitics of frontier AI.