Anthropic expects Mythos and Fable 5 back online 'in coming days'

When the White House ordered Anthropic to cut off foreign access to its most powerful AI models, the company responded by pulling the plug entirely, worldwide.

Now, Anthropic is telling users not to panic. Speaking at a press conference at the Conrad Seoul on Wednesday, Anthropic Managing Director of International Chris Ciauri said the company is very confident that access to its top-tier Mythos and Fable 5 models will be restored in the coming days.

For developers caught in the crossfire of national security orders and model shutdowns, the relief cannot come soon enough.

A disputed jailbreak and a sudden blackout

The worldwide shutdown followed an export control directive from the Trump administration citing cybersecurity risks. The order barred non-U.S. citizens from accessing Mythos and Fable 5, even if they were physically located inside the United States.

Behind the scenes, the mechanics were messy. Amazon, one of Anthropic's largest investors, was surprisingly among the parties that flagged a potential Fable vulnerability to officials.

Meanwhile, The Washington Post reported that access granted to a South Korean telecom company with suspected ties to China helped trigger the White House order. Semafor also reported lingering fears that China-linked groups may have accessed Mythos.

Because the mandate would have prohibited Anthropic's own foreign-national engineers from working with the models, the lab opted to suspend them globally. Anthropic pushed back hard, noting it was only presented with verbal evidence of a non-universal jailbreak that exists in rival models like GPT 5.5, adding that a narrow jailbreak should not justify pulling a commercial service.

Defending the frontier in Seoul

The press event in Seoul was originally supposed to celebrate Anthropic's commercial expansion in South Korea. Instead, reporters pressed executives on the export ban and Project Glasswing, an exclusive cybersecurity initiative with around 150 partners, including Google, Apple, Nvidia, Microsoft, Samsung Electronics, and SK Telecom.

Anthropic co-founder Tom Brown was scheduled to attend but was replaced at the last minute by Ciauri, who joined Anthropic Korea Representative Director Choi Ki-young on stage.

Mythos is notoriously powerful, capable of discovering flaws in code previously thought unbreakable, which is why Anthropic kept it restricted to Glasswing partners. Despite the tension, the executives pivoted to South Korea's strategic value, noting the country ranks 12th globally in usage with a rapidly expanding base of developers and designers.

Ciauri pledged deeper investment, confirming that Anthropic is actively hiring commercial, technical, policy, and operational staff in Seoul.

Why it matters

For enterprise teams and security researchers, this outage is a wake-up call. Cutting-edge AI models are no longer insulated developer tools; they are now treated as controlled digital munitions. If your mission-critical workflow relies on a single frontier model, you are now exposed to geopolitical crossfire that can pull your tools offline overnight.

For the AI landscape, this sets an aggressive precedent. The U.S. government is actively enforcing export controls on model access by nationality rather than simple geography, and it relies on rival tech giants to surface vulnerabilities. Labs aiming to deploy frontier-class intelligence internationally will face unprecedented compliance hurdles.

Anthropic's handling here was both principled and painful. Shutting down access for everyone rather than establishing internal second-class citizenship preserved company culture, but it rattled enterprise trust. Having lobbied extensively for strict AI guardrails, Anthropic is now learning that when government intervention finally arrives, it rarely plays by Silicon Valley's rules.