Uncle Sam Just Pulled the Plug: The World's Most Powerful AI Lasted 72 Hours -

Anthropic’s Fable 5 was supposed to be the future of AI, until it wasn’t. It begs the question, “How much is too much?”
It is a question that is largely unanswered as I write this post. The context is the U.S. government’s announcement restricting access to Anthropic’s Claude Fable 5 large language model.
It survived 72 hours before the government shut it off for everyone on Earth.

Here’s the rundown on the situation.

What Are Fable 5 and Mythos 5?

Think of these two models like a superhero and their secret government-only clone. Anthropic released Claude Fable 5 on June 9, 2026, as its most powerful AI model ever made available to the public. It’s built on the same underlying model as Claude Mythos 5, basically identical twins under the hood, but with one big difference: Fable 5 has a “bouncer” (safety classifiers) that blocks dangerous requests in cybersecurity, biology, and chemistry, automatically downgrading them to the weaker Claude Opus 4.8.platform.

Mythos 5, on the other hand, is the no-bouncer VIP version, same model with fewer restrictions, made available only to a tiny group of vetted government cyber-defenders and critical infrastructure operators through a secret program called Project Glasswing.

The Government’s Shutdown Order (June 12)

Here’s where it gets dramatic. Just three days after Fable 5 launched to millions of users, Anthropic received a government directive at 5:21 PM ET on June 12 from Commerce Secretary Howard Lutnick ordering it to block all foreign nationals from accessing both models. No Americans, no foreign nationals; not even non-citizen Anthropic employees.

The practical problem: you can’t easily verify who’s a foreign national on the internet. So Anthropic did the only technically feasible thing: it shut both models off for literally everyone. AWS confirmed it revoked access “for all users in all regions”. Fable 5 had a 72-hour lifespan.

Why Did the Government Do This?

The official reason: national security. Specifically, someone claimed they found a way to “jailbreak” Fable 5, essentially tricking it into doing the dangerous cybersecurity stuff Fable was supposed to block, accessing the Mythos-level capabilities that should have been gated off.

An unnamed company reportedly demonstrated this jailbreak to the administration, triggering alarm bells. Commerce Secretary Lutnick sent a letter to Anthropic CEO Dario Amodei saying the models would face export restrictions.axios+1

Anthropic pushed back — hard. The company said:

The “jailbreak” was narrow and non-universal, not a master skeleton key
It only exposed already-known minor vulnerabilities
Other public models like GPT-5.5 can produce the same capabilities
“Perfect jailbreak resistance is not currently possible for any model provider”
The shown technique “does not warrant the withdrawal of a commercial model used by millions”

Essentially, Anthropic is saying: “The government panicked over a tiny scratch and called it a mortal wound.”

The Bigger Political Context (It’s Complicated)

This isn’t happening in a vacuum. Anthropic and the Trump administration have been in a full-on feud since Anthropic refused to let the U.S. military use its AI for domestic surveillance and fully autonomous weapons systems. The Pentagon responded by putting Anthropic on a supply chain blacklist, deeming it too dangerous for government use.

So now Anthropic is simultaneously:

Too dangerous for the government (Pentagon blacklist) 🚫
Too dangerous for foreigners (Commerce export controls) 🌍🚫

This marks a historic first: it’s the first time the U.S. government has issued a recall-style export control on a deployed commercial AI model itself, rather than just restricting the chips or hardware that power AI. For years, export controls targeted Nvidia chips and semiconductor tools, never the AI model as the controlled object.

Even a former White House AI official, Dean Ball, noted this implies users will eventually need to prove their citizenship to use Anthropic models.

Legal Wrinkles

A U.S. judge has already determined that the Pentagon’s separate blacklist directive cannot be enforced, allowing government-adjacent organizations to keep using Anthropic services during ongoing litigation. That legal battle remains unresolved.

The NSA had actually authorized Mythos 5 for classified network use before the administration ordered the global shutdown — a contradiction that’s raising eyebrows.

Where Things Stand Right Now

As of June 13, 2026 (today):

Fable 5 and Mythos 5 are offline for all users worldwide
All other Anthropic models (Claude Opus 4.8, etc.) are unaffected
Anthropic says it’s “working to restore access as soon as possible” and promised more details within 24 hours
The government says the restriction could lift “within the next few weeks” once national security frameworks are reinforced
No confirmed timeline or compliance conditions have been publicly disclosedlinkedin+1

The precedent this sets is enormous. If a government can flip a switch and turn off a frontier AI model mid-conversation for everyone on Earth, the question of who really controls the most powerful AI systems just got a lot more complicated. 🌐

https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5

https://thehackernews.com/2026/06/anthropic-releases-claude-fable-5-its.html

https://www.cnbc.com/2026/06/09/anthropic-mythos-claude-fable-5.html

https://www.bleepingcomputer.com/news/security/us-gov-asks-anthropic-to-ban-foreign-national-access-to-fable-mythos

https://www.channelnewsasia.com/business/anthropic-united-states-block-advanced-ai-models-foreign-governments-6180561

https://www.reddit.com/r/artificial/comments/1u4ef3y/anthropic_suspends_access_to_claude_fable_and/