Just three days after its launch, Claude Fable 5 — Anthropic’s most powerful public model — was taken offline by a US government export control order. A week later, it still hasn’t come back. Alongside its unrestricted counterpart Mythos 5, the model was disabled for all users worldwide on June 12 after Amazon CEO Andy Jassy alerted senior officials to a jailbreak discovered by Amazon researchers. The shutdown has become a watershed moment for the AI industry, raising fundamental questions about model availability, national security, and the fragility of developer toolchains built on frontier models.
The Timeline
June 9 — Anthropic launches Claude Fable 5 to hundreds of millions of users and Claude Mythos 5 through Project Glasswing, a government-partnered program giving trusted cyberdefense organizations access to unrestricted AI. Developers immediately report that Fable 5 refuses 200 of 200 ProgramBench tasks, suggesting unusually aggressive safety restrictions. CEO Dario Amodei publishes a policy essay calling for government authority to block unsafe AI.
June 11–12 — Amazon CEO Andy Jassy alerts Treasury Secretary Scott Bessent and other officials about a Fable 5 jailbreak discovered by Amazon researchers. The technique, per Anthropic’s official statement, involved asking the model to review a codebase and identify software vulnerabilities — a capability Anthropic argues is available from other frontier models and used daily by cyber defenders.
June 12, ~5:21 PM ET — Commerce Secretary Howard Lutnick sends an export control letter to Dario Amodei, citing national security authorities. The order directs Anthropic to suspend access for all foreign nationals, including foreign national employees. Because Anthropic has no infrastructure to segment users by nationality, the models are disabled for everyone worldwide that evening. When Amodei tells Lutnick this means pulling the models globally, Bloomberg reports the Secretary’s reply: “That’s the point.”
June 13 — The Guardian and Reuters break the story. White House AI adviser David Sacks posts on X: “A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.” Anthropic disputes this, stating: “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”
June 14–15 — Anthropic dispatches technical staff to Washington DC for negotiations.
June 19 (today) — Both models remain offline with no resolution announced.
The Amazon Connection
The irony is sharp: Amazon is one of Anthropic’s largest investors and provides the cloud infrastructure powering Claude. Jassy’s warning — triggered by Amazon researchers legitimately testing Fable 5 — set off the chain of events that shut down the model. Fortune confirmed that “a warning from Amazon chief executive Andy Jassy, concerns about unauthorized Chinese access, and cybersecurity fears all reportedly led the Trump administration to cut off foreign access.” The episode has strained one of the most consequential partnerships in AI, raising questions about whether the relationship can withstand a regulatory crisis Amazon itself triggered.
Mythos 5 and Project Glasswing
Mythos 5 was the unrestricted version of Fable 5, with safety guardrails lifted for trusted partners via Project Glasswing. Priced at $10/M input tokens and $50/M output tokens, it was already proving its value in security research. The Wall Street Journal reported that Anthropic security researcher Nicholas Carlini — a well-known AI safety skeptic — used Mythos 5 to discover never-before-seen bugs in Ghost and Linux kernel code. He changed his stance after the experience: “It’s pretty clear to me that these current models are better vulnerability researchers than I am,” he told the WSJ, warning that the attacker-defender balance “seems like it’s probably coming to an end.”
Negotiations in Washington
Anthropic’s DC delegation includes Chief Compute Officer Tom Brown, Nicholas Carlini, External Affairs head Sarah Heck, and red-teaming lead Logan Graham. The Commerce Department has signaled willingness to let Fable 5 resume once Anthropic addresses the jailbreak — a technical fix the company believes is feasible. Mythos 5’s future is less certain, as its unrestricted capabilities raise harder questions about dual-use technology and export controls that may not have a quick resolution.
More than 100 cybersecurity experts signed an open letter arguing that removing Mythos 5 from the defensive toolkit makes everyone more vulnerable. The Council on Foreign Relations told the Financial Times: “Washington has just done more to boost Chinese AI models than Beijing ever could have hoped for.” Legal experts have also raised concerns — Charlie Bullock of the Institute for Law and AI called the legal theory “strange, very aggressive, and probably vulnerable to legal and constitutional challenge.”
What Developers Can Use Instead
For developers left scrambling after losing access mid-project, several alternatives are available:
- Claude Opus 4.8: Anthropic’s previous flagship remains available with strong coding performance, albeit below Fable 5’s capabilities.
- OpenAI GPT-5.5: Anthropic itself acknowledged the jailbreak capability “is widely available from other models (including OpenAI’s GPT-5.5).” GPT-5.5 is fully available and offers comparable software analysis abilities.
- GLM-5.2: Zhipu AI’s latest model offers competitive coding performance outside US export control scope, making it a geopolitically neutral option.
- Claude Sonnet 4.6: A lighter, faster Anthropic model for everyday coding tasks, though significantly less capable than Fable 5.
The episode has also accelerated interest in locally-run and open-weight models, as developers confront the fragility of relying on hosted frontier models that can disappear overnight.
The Bigger Picture
The Claude Fable 5 shutdown sets a new precedent: a frontier AI model taken offline not by technical failure or business decision, but by a government asserting national security authority over a capability that did not exist six months ago. Whether Fable 5 returns — and in what form — will be determined in Washington, not San Francisco. As one Project Glasswing partner told the Washington Post: “We knew this technology was powerful. We didn’t know it could be taken away this fast.”
Deutsche Bank’s Jim Reid captured the commercial lesson bluntly: “You can’t rely on something that could be switched off.” For the developers who lost their primary coding tool on June 12, that lesson is already painfully obvious.
This article was prepared using primary sources including Anthropic’s official statement on the export control directive (anthropic.com/news/fable-mythos-access), the company’s Fable 5/Mythos 5 launch announcement, and reporting from The Guardian, Forbes, Fortune, Reuters, the Wall Street Journal, and Snyk. All sources verified as of June 19, 2026.