Fable 5: Inside the Most Powerful AI Model Ever Exported and Export Banned Overnight

What Is Fable 5?

Fable 5 is Anthropic’s flagship AI model, released on June 9, 2026, alongside a more capable sibling, Mythos 5. In a blog post, Anthropic wrote Fable 5 is “a significant improvement” in general capabilities over any model the company has previously made publicly available, with state-of-the-art performance across software engineering, knowledge work, vision, and scientific research.

In terms of innovation, Fable 5’s particular distinction was in stamina, or handling long-running, multi-stage, and asynchronous workflows. This includes complex code refactoring and research synthesis that require a model to plan its approach, check its progress against a desired goal, and refine its work as it goes rather than wait for the next prompt.

The real-world impact was impressive. Early user testing by Stripe found Fable 5 could complete software engineering workloads in a matter of days that would normally take a full team months. This included a codebase-wide migration across a 50 million-line Ruby codebase in one day that would typically have taken more than two months. One finance team scored the model highest of any AI system they tested against on senior-level reasoning benchmark, while one trading firm noted Fable 5 aced a multi-part evaluation covering factual lookup, root-cause analysis, and expected-value reasoning.

As is typical with Anthropic models, this level of power comes with built-in guardrails.

Anthropic launched Fable 5 with safeguards in which some queries would instead be directed to its next-most-capable public model, Claude Opus 4.8 meaning unsafeguarded capabilities in some areas, like cybersecurity, could not be misused to cause harm. A more powerful, lightly restricted Fable 5 counterpart called Mythos 5 was reserved for a small group of vetted cyberdefenders deployed through a government partnership called Project Glasswing.

The Jailbreak That Led to an Export Control Directive

Things went sideways three days after launch. On June 12, administration officials announced Amazon AI experts were able to orchestrate a jailbreak on Fable 5 within hours of its release, finding a workaround that bypassed Anthropic’s stated restrictions to unlock the model’s full cyber capabilities.

What happened next was historic. The US government issued an export control directive to suspend all access to Fable 5 and Mythos 5 for any foreign national, whether based in the United States or not, including Anthropic’s own foreign national employees. Because Anthropic’s system could not distinguish between domestic and international users at the infrastructure level, Anthropic said it had to disable both models for all customers worldwide to comply.

The legal justification was just as unusual as the action. The June 12 order invoked the “deemed export” doctrine, a rule that was written for physical items like semiconductor blueprints but treats giving controlled technology to a foreign national inside the US as an equivalent of directly exporting it to that person’s home country. At the time of the ban, the doctrine had never before been applied to real-time AI inference over a cloud endpoint.

Anthropic pushed back on how serious the demonstrated flaw was. Anthropic said it had reviewed the jailbreak technique in detail and that it exposed only a small number of known, relatively straightforward, minor vulnerabilities that other publicly available models could also be made to expose. Anthropic maintained that nobody had found a “universal jailbreak” that could broadly bypass Fable 5’s safeguards and that its “defense-in-depth” approach limited the model’s potential risk to roughly the same level as other models already deployed by companies across the industry.

Tensions Behind the Scenes

The export ban was the culmination of a months-long escalation between Anthropic and the Trump administration. Behind the scenes, it was rooted in a fundamental disagreement over military use of AI.

According to a report, Anthropic reached a deal with the Pentagon in July 2025 that would have made Claude the first frontier model approved for use on classified government networks, a major trust milestone no other AI lab had achieved at the time. That deal fell apart in February 2026 when the Pentagon demanded Anthropic allow military use of Claude for “all lawful purposes,” including in lethal autonomous weapons and mass surveillance of Americans, a demand Anthropic refused.

Anthropic’s refusal had consequences. The Pentagon blacklisted the company as a result of the dispute, effectively deeming the company too dangerous for government use and with the export controls, too dangerous for foreign use as well. Defense Secretary Pete Hegseth made the administration’s posture explicit in a post, writing that the Department of War had banned Anthropic from its building months earlier and that “every passing day proves why that was the right move.” Fable 5 has also had direct human costs that became a flashpoint. Andrej Karpathy, one of Anthropic’s own top AI researchers, was unable to access Fable 5 and Mythos 5 because he isn’t a US citizen, an image that came to symbolize the blunt edge of the order.

Foreign National Allies Left Out in the Cold

Because the order applied to all foreign nationals regardless of nationality, the ban had a disproportionate impact on countries that consider themselves close US allies.

The ban applied as much to countries with intelligence-sharing and mutual defense agreements with the US as to known adversarial nations like China and Russia, marking a departure from how the US had traditionally used export controls. French President Emmanuel Macron called the move a “wake-up call” on AI risk but also a “bad thing” that would “push back” on its own interests, while also framing it as a strictly nationalist reaction.

Canadian Prime Minister Mark Carney said the Fable and Mythos situation was a lesson in what happens when countries over-rely on a single foreign supplier, arguing that allies needed to build out and diversify rather than just accept the disruption.

As of this writing, it remains unclear when or if Fable 5 will be restored, though Anthropic has said that both models will be returned “in the coming days.” Eight days after the ban took effect, Trump reportedly told reporters that he no longer viewed Anthropic as a national security risk after a meeting with CEO Dario Amodei at the G7 summit in Évian-les-Bains.

He described Amodei as “nice” and “smart,” and an Anthropic spokesperson said the company was grateful for the administration’s “ongoing partnership” in trying to resolve the situation. Still, Anthropic’s international managing director said at the company’s Seoul office opening that he was “very confident” both models would return “in the coming days,” though no official restoration date has been set.

In a global sense, the episode has reignited a wider conversation around AI sovereignty and interest in homegrown models. As one analyst told Al Jazeera, the controversy has drawn fresh attention to Mistral, the Paris-based startup widely viewed as the EU’s only major homegrown frontier-model competitor.

What to Watch

While the outcome is still to be determined, the Fable 5 story has already set new precedents: for the first time, a government has used export control law to restrict access to an AI model itself, not just the chips that power it. It’s a reminder that as frontier AI systems grow more capable, the rules governing who can use them and who decides are still being written in real time, and often during the week a new model launches.

For businesses and developers who had already started building on Fable 5’s expanded context window and long-horizon reasoning, the sudden outage has been a painful reminder of platform risk. For policymakers, it’s a preview of the kind of high-stakes, live-fast decisions that capability growth is going to keep forcing in AI, with global ramifications hinging on a single letter from a single government agency.