US Government Orders Anthropic to Suspend Fable 5 and Mythos 5: What Happened

On June 12, 2026, Anthropic announced that the United States government had issued a directive forcing it to suspend access to two of its frontier AI models, Fable 5 and Mythos 5. The order arrived at 5:21pm ET and cited national security authorities under export control law. Within hours, both models went dark for every Anthropic customer. This article breaks down exactly what happened, which models are affected, why the government acted, how Anthropic responded, and what it means if you rely on these tools.

What exactly did the US government order?

The government issued an export control directive — a legal instrument that restricts who can access a given technology — instructing Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether located inside or outside the United States. Critically, the restriction even covers Anthropic's own foreign-national employees.

Because Anthropic cannot practically separate foreign nationals from the rest of its global user base in real time, the only way to guarantee compliance was to disable both models entirely, for everyone. The company was explicit that this was done "to ensure compliance," not because it believes every user poses a risk.

Export control directive

A legally binding government order that limits access to a technology on national security grounds, often restricting use by foreign nationals.

Jailbreak

A technique that bypasses an AI model's safety guardrails to make it produce restricted or prohibited output.

Universal jailbreak

A single method that broadly defeats a model's safeguards across many tasks — the most serious category. No tester has found one for Fable 5.

Non-universal jailbreak

A narrow bypass that only works in specific, limited circumstances and yields little to no additional capability.

What are Fable 5 and Mythos 5?

Fable 5 and Mythos 5 are Anthropic's newest frontier models, sitting at the top of its lineup alongside the long-standing Opus, Sonnet, and Haiku families. They were positioned as the company's most capable systems for complex reasoning, coding, and agentic tasks. Because they represent the cutting edge of capability, they also carry the strictest safety treatment — which is exactly why a jailbreak claim against Fable 5 drew government attention rather than being treated as a routine bug report.

When Anthropic launched Fable 5, it published a detailed safety blog post describing the guardrails it built specifically to reduce misuse in sensitive areas such as cybersecurity. The company has consistently framed these two models as the place where it invests the most in safety testing, monitoring, and data-retention policies — context that matters for understanding why the current dispute is happening over these models and not the others.

Which Anthropic models are affected?

Only the two newest frontier models are suspended. The rest of Anthropic's lineup continues to operate normally.

If your workflow runs on Opus, Sonnet, or Haiku — including most Claude and Claude Code usage — you are not affected by this directive and can continue working as normal.

Why did the government act? The jailbreak concern

The directive did not spell out the specific national security concern. Anthropic's understanding is that the government believes someone discovered a method of bypassing, or "jailbreaking," Fable 5. The company reviewed a demonstration of the technique and reported that it surfaced only a small number of previously known, minor vulnerabilities.

According to Anthropic, the potential jailbreak essentially consists of asking the model to read a specific codebase and fix any software flaws it finds — a routine task that defenders and developers perform every day. The company validated that the same level of capability is "widely available from other models, including OpenAI's GPT-5.5," and does not require a bypass at all.

Timeline of events

How did Anthropic respond?

Anthropic is complying with the legal order while publicly pushing back on its justification. The company laid out its safety posture for Fable 5, which it summarizes as a "defense in depth" strategy:

• Strong safeguards that, by the company's account, are so strict that many users complained they were overly broad.
• Thousands of hours of red-teaming before launch, involving the US government, the UK AI Safety Institute, third-party organizations, and internal teams.
• No universal jailbreak found to date — no method that broadly unlocks a wide range of cyber capabilities.
• 30-day data retention on Fable usage — a costly policy adopted specifically to research and quickly mitigate any jailbreaks.
• Monitoring designed to detect and shut down successful attacks rapidly.

Anthropic argues that perfect jailbreak resistance is not currently possible for any provider, and that recalling a model deployed to hundreds of millions of people over a narrow, non-universal jailbreak would, if applied industry-wide, "essentially halt all new model deployments for all frontier model providers."

What does this mean for you?

The practical impact depends entirely on which models you use:

• If you use Fable 5 or Mythos 5: access is currently unavailable. Switch critical workflows to Opus, Sonnet, or Haiku until access is restored.
• If you use Opus, Sonnet, or Haiku: nothing changes — these models are explicitly unaffected.
• If you build products on Anthropic's API: audit which model IDs your application calls. Anything pinned to Fable 5 or Mythos 5 will fail until the suspension lifts; everything else continues to work.

Anthropic has stated it believes the order rests on a misunderstanding and is working to restore access "as soon as possible," but it has not committed to a specific timeline.

The bigger picture: government power over AI deployments

Anthropic has publicly supported the idea that the government should be able to block genuinely unsafe deployments — but only "as part of a statutory process that is transparent, fair, clear, and grounded in technical facts." The company argues this particular action does not meet those principles, because it was based on a narrow, verbally communicated finding rather than a transparent technical process.

The episode is an early, high-profile test of how export control authority applies to commercial AI models, and the precedent it sets could shape how quickly frontier models reach the market in the future.

How to check whether you're affected

If you are unsure whether this suspension touches your work, a quick checklist helps:

1. Identify your model. Open your tool or API configuration and find the exact model name. If it is anything other than Fable 5 or Mythos 5, you are not affected.
2. Check API integrations. Applications often pin a specific model ID. Search your codebase or environment variables for references to Fable 5 or Mythos 5 and prepare a fallback to Opus or Sonnet.
3. Set a temporary fallback. Where possible, route requests to Opus, Sonnet, or Haiku so your product keeps running even while the two models are unavailable.
4. Watch official channels. Anthropic committed to sharing more details and is working to restore access, so its status page and announcements are the authoritative source for when the models return.

For most everyday Claude and Claude Code users, no action is required, because those experiences run on models that remain available. The users most affected are those who deliberately selected Fable 5 or Mythos 5 for their highest-end reasoning or coding tasks.

Why this matters beyond Anthropic

Whatever the final outcome, this is one of the clearest examples yet of a government using export control authority to pull a widely deployed commercial AI model offline on short notice. How it is resolved — quickly and transparently, or slowly and opaquely — will signal to every frontier lab how much regulatory certainty they can count on when they ship their most capable systems. For businesses building on top of these models, it is also a reminder to design for model portability, so that a single provider or model going dark does not take critical workflows down with it.