In less than three weeks, Anthropic’s most advanced AI models went from global darlings to national-security risks—and back again—after an abrupt Trump administration crackdown triggered a scramble over how to police cutting‑edge AI.

Early June: Hype around Mythos 5 and consumer model Fable 5

Anthropic had just finished hyping its new Claude Fable 5 and Claude Mythos 5 models, with Mythos framed as a powerful cybersecurity tool and Fable as the safer, consumer-facing sibling built from the same core technology.¹ Mythos had already been limited to select organizations under “Project Glasswing” because it was “too adept at finding and exploiting cybersecurity vulnerabilities.”²

June 12: Sudden export controls and a global shutdown

On Friday, June 12, the U.S. Commerce Department placed export controls on both models, requiring Anthropic to block foreign nationals—even its own employees—from access.³ Lacking any practical way to verify nationality in real time, Anthropic “abruptly disable[d]” access to Fable 5 and Mythos 5 for everyone worldwide.²

Officials feared a specific “jailbreak” that could unlock dangerous capabilities, reportedly first flagged by Amazon researchers, pushing the administration to act with an evening ultimatum.⁴ Anthropic argued the order stemmed from a misunderstanding of that jailbreak and flew executives to Washington to negotiate with the White House.²

Late June: Negotiations, pressure, and partial reopening

Over the following weeks, Commerce Secretary Howard Lutnick and Anthropic hammered out new safeguards. Lutnick said Anthropic agreed to “proactively detect and address security risks” and work on protocols and standards for Mythos, Fable, and future models.⁵ Meanwhile, U.S. access to Mythos 5 quietly resumed for a select group of government‑approved organizations, mirroring restrictions on OpenAI’s latest models.⁶

Behind the scenes, Anthropic was also grappling with its broader fraught relationship with the Trump administration, including an ongoing lawsuit over being labeled a Defense Department “supply chain risk,” all while preparing for a possible IPO following a confidential S‑1 filing on June 1.²

June 30–July 1: Full deal and global redeployment

On June 30, Commerce informed Anthropic that export controls on Claude Fable 5 and Mythos 5 were lifted, and that it would “no longer need a license for exports or in-country transfers” of the models.⁷ Anthropic announced it would begin restoring access on July 1, first on its own platforms and then through AWS, Google Cloud, and Microsoft’s Foundry.⁴

To get there, the company trained an “improved safety classifier” targeting the jailbreak; in most reports, that safeguard blocks the technique in the vast majority of cases, with any remaining outputs limited to already‑known or patched vulnerabilities.⁴⁶ Anthropic also committed to 24/7 monitoring for new jailbreak threats and structured red‑teaming with outside hackers.⁷

Competing narratives: Safety win or political leverage?

The Trump administration presented the episode as a model of aggressive, America First AI governance, with senior officials boasting that government and industry “worked together in a way we have never seen before” to deploy “the best tech … as quickly and safely as possible.”⁶ Supporters said the pause forced real safety testing and tighter collaboration.

Anthropic emphasized its gratitude to users and partners and framed the outcome as a chance to prototype pre‑release testing, rapid information‑sharing, and shared security standards with government.¹⁴ The company noted it had already been voluntarily committing to many of the new obligations before the export rule, a point cybersecurity experts seized on to argue the harsh curbs looked less like a precise security fix and more like political leverage over a sometimes‑critical AI lab.⁵

Yet some in industry and government agreed on one thing: with rival Asian models like Fugu and Tulonfeng approaching Mythos‑level capabilities, keeping U.S. frontier models locked down indefinitely risked ceding the lead in AI.⁵ The final compromise leaves Washington with the right to re‑impose curbs at any time, while Anthropic’s most advanced systems return to a global market now watching closely how far future AI crackdowns might go.⁷