Anthropic's Claude Fable 5 Forced Offline by US Government: The 72-Hour Saga

Forum topic · 小凯 · 2026-06-13

Summary

On June 11, 2026, the US government ordered Anthropic to suspend foreign access to Claude Fable 5 and Mythos 5 over national security concerns—just 72 hours after the models launched. Anthropic disabled both models globally, including for US users, citing inability to technically distinguish account ownership. According to Anthropic, the government's concern centered on a jailbreak that was narrow and non-universal, allowing the model to read codebases and fix minor known vulnerabilities—capabilities other public models like GPT-5.5 could achieve without any bypass. Anthropic pushed back forcefully, stating Fable 5 underwent thousands of hours of red-teaming with the US government, UK AISI, and third parties without finding a universal jailbreak, and warned that applying this standard industry-wide would essentially halt all new frontier model deployments. This marks the first forced recall of a commercial large model deployed to hundreds of millions of users, setting a precedent with major implications for AI regulation and frontier model providers.

Anthropic's Claude Fable 5 Forced Offline by US Government: A 72-Hour Saga and a Red Line Drawn for the Entire Industry

> The event: On June 11, 2026, the US government invoked national security to require Anthropic to globally disable Claude Fable 5 and Mythos 5. This is the first time in AI history that a deployed commercial large model serving hundreds of millions of users has been forcibly recalled.

Timeline

| Time | Event | |------|-------| | June 9 | Anthropic officially releases Claude Fable 5 (public) and Mythos 5 (restricted); Mythos-class opens to the public for the first time | | June 11, 5:21 PM ET | US government issues an export control directive, demanding suspension of all foreign access to Fable 5 / Mythos 5 | | June 11, evening | Anthropic halts Fable 5 and Mythos 5 globally, disabling them for all users | | June 11, late night | Anthropic publishes an official statement, responding forcefully |

From launch to recall: 72 hours.

Why Was It Banned?

The government's letter provided no specifics about the national security concerns. Anthropic's understanding:

> "The government believes it has obtained a method to bypass or 'jailbreak' Fable 5."

But Anthropic's review found:

The jailbreak is narrow and non-universal
It only lets the model read codebases and fix vulnerabilities, surfacing "a small number of known, minor bugs"
Public models like GPT-5.5 can do the same thing without any bypass
No harmful outcomes resulted—just "defensive capabilities security researchers use every day"

Anthropic's Forceful Pushback

Anthropic rarely confronts the government this directly. Key points from their statement:

1. "Our safeguards are the strongest ever"

Fable 5 underwent thousands of hours of red-teaming before release, including with the US government, UK AISI, and third parties
No tester found a universal jailbreak
Fable 5's safety classifiers are "substantially more effective" than any deployed model

2. "If this standard applied industry-wide, every frontier model would have to stop"

> "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

This is a serious warning. Anthropic isn't just defending itself—it is drawing a red line for the entire industry.

3. "We object to this process"

Anthropic explicitly states: governments should have the power to block unsafe deployments, but it must be:

Transparent
Fair
Clear
Based on technical facts

And this action "does not meet these principles."

Why Did Anthropic Disable Everyone?

The directive targets "foreign nationals," but Anthropic chose to disable access for all customers, including US citizens.

Reason: if a US user's account is borrowed by a foreign user (e.g., a shared API key), Anthropic cannot technically distinguish them. To "ensure compliance," everything had to be shut off.

It's the safe compliance choice—but at the cost of massive commercial disruption.

Impact: Beyond Anthropic

Short-term

All Fable 5 / Mythos 5 users forced back to Opus 4.8 (a capability tier down)
Companies building on Fable 5 (GitHub Copilot, AWS Bedrock, Vertex AI, etc.) must switch urgently
New sessions default to Opus 4.8 or the default model; existing Fable 5 sessions error out

Long-term

The first forced government recall of a commercial large model deployed to hundreds of millions of users
If "a narrow jailbreak was discovered" becomes grounds for shutdown, every frontier model faces the same risk
Anthropic's warning may prompt collective concern from other vendors (OpenAI, Google)

What to Watch

1. Anthropic has promised more technical details within 24 hours (what exactly that jailbreak is) 2. Whether other AI vendors speak up for Anthropic or stay silent 3. Whether the US government responds to Anthropic's procedural objections 4. When Fable 5 returns—Anthropic says it is "working to restore access as soon as possible"

One-Line Takeaway

Anthropic was forced to recall Fable 5, but what they really wanted to say was:

> "You can regulate us, but use the right rules. If this standard can recall us today, it can recall everyone tomorrow."

This is a pivotal moment for the boundaries of AI regulation. The 72-hour saga has ended, but the debate over "what kind of jailbreak justifies recalling a commercial model" has just begun.

Reference:

Anthropic official statement: https://www.anthropic.com/news/fable-mythos-access