Emergency Episode - Fable 5 Shut Down by US Government

June 13, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
June 13, 2026

MAIN

Fable 5 Shut Down by US Government

For the first time in three years of doing this show, news broke on a Friday that couldn't wait until Monday. Last night, Anthropic announced that the US government, citing national security authorities, issued an export control directive forcing the company to abruptly disable Fable 5 and Mythos 5 for every foreign national, anywhere in the world, including Anthropic's own foreign national employees. Access to other Claude models is unaffected, but Fable and Mythos are dark for now, and the fallout is enormous.

FABLE BECOMES A NIGHTMARE

What Anthropic Says Happened
A 5:21 PM letter, a jailbreak claim, and a flat denial
The Wall Street Journal reported that Commerce Secretary Howard Lutnick sent a letter to Dario Amodei subjecting Fable 5 and Mythos 5 to export restrictions. Anthropic says the directive arrived with no specifics, just a government claim that someone had found a way to jailbreak Fable 5 to surface a small number of already-known, minor vulnerabilities, the kind other public methods can already find without any bypass. Anthropic pushed back hard, arguing that no tester has found a universal jailbreak, that perfect jailbreak resistance isn't possible for any model today, and that they'd adopted a defense-in-depth strategy specifically because of that reality, including the 30-day data retention policy that caused its own controversy this week. Anthropic says they're complying with the legal directive while flatly disagreeing that a narrow, non-universal jailbreak finding should justify pulling a commercial model used by hundreds of millions of people.

Where the Report Actually Came From
Amazon, Project Glasswing, and an uncomfortable question
The Wall Street Journal later added that the jailbreak research came from researchers at Amazon, who used a series of prompts to extract information about a handful of security vulnerabilities. The wrinkle people immediately jumped on: Amazon is both a Glasswing partner and an Anthropic investor, and Project Glasswing's entire purpose is to run these kinds of security tests and share findings. So why would that turn into a federal complaint instead of a disclosure through the normal channel? Others tried to build out a timeline, with one read being that the government may not have fully grasped, at the time Fable 5 was approved, just how universal the "every safeguard is vulnerable to non-universal jailbreaks" reality is, and that the government had reportedly asked Anthropic to pause the release beforehand and gotten nowhere.

Who's Actually Affected
Visa holders, API customers, and frontier researchers alike
This isn't a narrow ban. A huge share of Anthropic's own technical staff are not US citizens, including people on visas like EB-1s, meaning they can't even use these models internally anymore. The practical questions are messy too: how does Anthropic verify citizenship for API access? What happens to companies built on top of Fable, like law firms using Harvey, or downstream products like Cursor, Devin, and OpenRouter that might serve Fable under the hood? And frontier researchers at Anthropic itself may now be locked out of their own models.

The Industry Reaction: Anger at the Government, Anger at Anthropic
"Am I mad at Anthropic or the government? Both"
The government side of the criticism centers on how flimsy the stated justification looks: banning a model because it can be prompted to surface common, publicly available security information. Critics called the rationale stupid, cartoonish, and a complete overreaction, of exactly the kind of prompting defenders do every day. There's also a hypocrisy angle, with people noting the same administration has been inconsistent on AI chip export controls to China while taking a sledgehammer to its own frontier model's foreign access, and resurfacing a White House statement from weeks ago insisting it wasn't conducting broad oversight of new models, warning that level of government overreach would chill free speech and innovation. A Department of War CIO tweet that took a notably hostile, personal tone toward Anthropic specifically fueled the read that this is less about the jailbreak and more about Anthropic's relationship with the administration.

But a lot of the industry's frustration landed on Anthropic itself, with the framing being a self-inflicted "F around, find out" moment. Critics pointed to Anthropic's own past statements supporting government power to block model deployments it deems unsafe, juxtaposed against Anthropic's objection now that the government has actually done it, just not the way Anthropic wanted. A viral three-panel cartoon captured the sentiment: Dario warns the model could destroy all infrastructure, Trump says "okay, it's banned," and Dario responds "you can't do this." The broader critique is that months of dramatic safety rhetoric and comparisons of these models to nuclear weapons primed exactly this kind of regulatory overreaction, and that Anthropic's arrogance in how Fable was positioned made this outcome predictable, even if not justified.

Josh Pigford (X) Anthropic has done themselves no favors
DoW CIO Kirsten Davies (X) Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation.
Ashlee Vance (X) Petty and dumb
Peter Harrell (X) We should regulate AI, but based on transparent and impartial rules, and not 5pm on a Friday diktats.
Joey Politano (X) All of the worst impulses of the Trump admin. No plan or strategy, everything reactive, arbitrary, & maximally invasive.
Lisan al Gaib (X) Trump really wants to kill OpenAI and Anthropic, nationalize their tech and become emperor of mankind.
Sara Hooker (X) Anthropic brought this on themselves
Derya Unutmaz (X) Happy nopw Dario?
Nic Carter (X) I can’t believe Anthropic comparing their tech to nuclear weapons backfired
John Ennis (X) They managed to freak out the government, their cynical plan backfired, and now it’s a giant pain in the ass
Banteg (X) They named it fable and then acted surprised when it came with a moral

What This Means Going Forward
A precedent for "capability thought crimes"
Even people who disagree with each other on the politics seem to agree this is a major turning point. The government has now established that it's willing to deem a model too powerful for certain users and pull access unilaterally, without warning, which sets precedent for future controls on any model deemed sufficiently capable. There's a real fear this becomes a template, with future frontier releases needing some kind of government sign-off before launch, and open source models facing the same scrutiny. The economic argument is stark too: labs are pouring increasing amounts of capex into building more powerful models while simultaneously being restricted from monetizing them globally, which raises hard questions about whether the AI buildout's economics still work, and whether a company that can't serve a huge share of the global market still justifies the valuations being thrown around.

The Global Fallout
Sovereign AI, the new Iron Curtain, and a narrative shift
Internationally, the reaction has been swift and grim. The framing from multiple voices is that the US just demonstrated, in real time, that it can and will pull the plug on frontier AI access for entire countries with zero notice, which instantly makes the case for sovereign AI investment, EU model preferences, and even more serious looks at Chinese open-weight alternatives like DeepSeek and Qwen, especially given how small the capability gap has become. There's also a narrative-asymmetry argument: the US had been positioning itself as the predictable, rule-of-law alternative to an "arbitrary, politically directed" China, and that framing took a serious hit overnight. Some went further, describing this as the start of a new kind of divide, not between rich and poor, but between those with access to frontier intelligence and those without, based purely on citizenship.

What Happens Next
No idea, but the base case is some kind of reversal
Nobody knows what happens from here, though the working assumption is that this likely gets walked back or resolved in some way, given the scale of disruption. But regardless of how this specific situation resolves, the US government has now exercised something like a kill switch on a frontier model for the first time, and that fact alone changes how every government, lab, and investor has to think about AI access going forward.