Anthropic Launched Its Most Powerful AI Ever. The US Government Pulled It Three Days Later.

On June 9, 2026, Anthropic launched Claude Fable 5;  the most capable AI model the company has ever released. It sat above the Opus class in a new tier called Mythos. It had a one-million-token context window, 128,000 output tokens, and always-on adaptive thinking: a mode that makes the model automatically double-check its own reasoning before responding, every single time.

It was available for exactly seventy-two hours.

On June 12, the US government issued an export control directive ordering Anthropic to suspend all access to Fable 5 and its companion model, Mythos 5. The stated reason was the discovery of a potential jailbreak method. Anthropic complied and the models went dark. It is the first time a government has ever used export controls to pull a commercial AI model from the market.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Major Highlights

• Claude Fable 5 and Mythos 5 : Anthropic’s first Mythos-class models, representing a new tier above Opus, launched on June 9, 2026, available on the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry the same day.
• The models featured a 1-million-token context window, 128K output tokens, and always-on adaptive thinking, which basically means a reasoning mode that cannot be disabled, built to make the model catch its own errors in real time.
• Safety classifiers were included that automatically fall back to Claude Opus 4.8 for high-sensitivity domains such as cybersecurity and biology.
• On June 12, three days after launch, the US government issued an export control directive requiring Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including Anthropic’s own non-US employees.
• Because enforcing the directive only for foreign nationals was practically impossible without disabling the models entirely, Anthropic suspended access for all users globally.
• The government’s concern: A method of bypassing Fable 5’s safety guardrails; a jailbreak had been discovered and demonstrated to officials.
• Anthropic publicly disagreed with the decision, stating their review of the jailbreak demonstration found it exposed only a small number of previously known, minor vulnerabilities, and that pulling a model used by hundreds of millions of people was a disproportionate response.
• Anthropic says it is working to restore access and is engaged with the government to resolve the situation.
• All other Anthropic models — Claude Opus 4.8, Sonnet, Haiku — remain fully available and unaffected
• This is the first-ever government-ordered suspension of a commercial AI model using export control authority.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

KINI BIG DEAL

What the US government did was a safety move. A jailbreak was demonstrated on the world’s most powerful AI model; one with a one-million-token context window and always-on reasoning. Pulling it first and asking questions later is arguably the correct instinct. You do not leave a compromised system running while you investigate how bad the breach is.

The problem is not that they acted. The problem is what acting looks like from everywhere else.

From Lagos, this is what it looks like: the model was available on Monday. By Thursday, it was gone without a warning. No process for governments, universities, hospitals, or businesses outside the United States to be part of the conversation, or even to receive advance notice. A safety decision made in Washington became an immediate operational reality for hundreds of millions of people who had no input and no recourse.

If a jailbreak is serious enough to pull a frontier AI model, then the safety of these tools is a global concern. But the governance of that safety is not global. It happens in one country, under one regulatory framework, with one set of interests at the centre. The rest of the world participates as users, not as stakeholders.

That is what this week made visible. Fable 5 will likely return, but the structure that made this possible — where a safety call in Washington is automatically a shutdown for everyone else — that does not go away when it does.