US Eases Export Ban on Anthropic's 'Mythos' AI After Standoff

The US government has blinked. Just weeks after abruptly forcing Anthropic to pull the plug on its most powerful AI models, the Commerce Department has reversed course. It ends a tense standoff that sent shockwaves through the AI industry. This partial lifting of **Anthropic Mythos export restrictions** is a pivotal moment, establishing a new—though fragile—framework for managing access to technology deemed a national security risk.

It all started on June 12. Citing national security, the government ordered Anthropic to suspend all access to its new Mythos and Fable 5 models for any foreign national. Anywhere. The order was so sweeping that Anthropic said its only option was to shut down the models for *everyone*. Globally. But after two-and-a-half weeks of intense collaboration, things have changed. Fable 5 is back for all. Mythos, the more potent of the two, is being released on a tight leash to a select group of approved US-based organizations.

Commerce Secretary Howard Lutnick announced the move on social media. "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." So what's the deal? According to a letter from Lutnick reviewed by multiple outlets, the agreement hinges on Anthropic's promise to “proactively detect and address security risks" and help the government forge standards for future models.

What Spooked the Government?

So what spooked the government? At its core, the alarm was about the unprecedented cybersecurity powers of Mythos and its sibling, Fable 5. These are not your average chatbot updates. Far from it. Anthropic’s own documents and security analyses show that Mythos can autonomously find 'zero-day' software vulnerabilities, write working exploits for them, and even chain minor bugs together to launch sophisticated cyberattacks. This represents a huge leap in agentic AI—systems that can perform complex, multi-step tasks with little to no human help.

The trigger? Reports say it was a demonstration by Amazon researchers. They showed how to bypass Fable 5's safeguards, letting it identify and even detail how to exploit software bugs. Anthropic downplayed it, saying the technique only found known, minor issues. But Washington wasn't convinced. The demonstration was enough to prove that the technology, in the wrong hands, posed a grave threat. The fear was crystal clear: an adversary could use the model to unleash devastating cyber-attacks on critical infrastructure.

This put Anthropic in a tough spot. Here's a company founded by ex-OpenAI researchers specifically to prioritize AI safety. Their entire philosophy is built around ideas like 'Constitutional AI', a complex framework designed to align models with ethical principles. And yet, that high-minded mission seemed to clash with the raw power they'd just unleashed. The government's order forced a company obsessed with safety to stare down the dual-use problem of its own creation.

A New Template for AI Governance

This whole episode is more than just corporate drama. It's a live-fire stress test for governing frontier AI.

And the resolution? A managed, tiered-access system. This is fast becoming the default playbook for any tech deemed too powerful for a wide-open release. We've seen this before. It's a striking parallel to rival OpenAI, which launched its GPT-5.6 model under a similar government-approved partner program.

The teamwork between Anthropic and the US Department of Commerce—specifically its Center for AI Standards and Innovation (CAISI)—points to a much deeper entanglement between Silicon Valley and Washington. Welcome to the messy reality of self-regulation meeting state power. Commerce Secretary Lutnick's letter confirmed that Anthropic will help the government write future protocols, turning a frantic, ad-hoc response into a potential blueprint. This all fits into a broader push for voluntary model release standards industry-wide, a critical move as the world grapples with increasingly autonomous systems.

But the initial ban revealed deep geopolitical fractures. Of course it did. US allies in Europe and Asia were frustrated, suddenly dependent on decisions made in D.C. It was a stark reminder of the global reach of American AI policy. The whole affair throws a spotlight on the fundamental tension in the never-ending battle between open and closed AI. Proponents of openness want broad access to spur innovation. But governments? They're leaning hard toward containment when national security is at stake—a fear amplified by a recent UN report on unchecked AI risks.

What Comes Next?

The immediate crisis is over. Fable 5 is back online worldwide, and Mythos is in the hands of a vetted list of over 100 US groups. But this story is just getting started. Anthropic has promised to work even more closely with the government, giving them pre-release access to future models and dedicating teams to joint safety research. The company is also huddling with partners like Amazon, Microsoft, and Google to build a common framework for judging AI 'jailbreaks.' It’s a clear sign the industry knows it needs to get on the same page.

But the government's leash is still tight. Lutnick's letter makes it plain: the administration can slap the restrictions right back on if things go south or if Anthropic doesn't hold up its end of the bargain. One of the world's top AI labs is, effectively, on probation.

The Anthropic-Mythos affair has set a precedent, and it's a big one. Gone are the days of dropping a frontier AI model on the world with just a blog post and an API key. That era is over. The new normal is cautious, state-sanctioned rollouts, where national security slams the brakes on pure technological acceleration. The real question is whether this fragile truce between innovation and control can possibly hold as the tech gets unimaginably more powerful.