FOLLOW US

Why Anthropic's AI Got Banned and What It Reveals About America's Broken Approach to AI Regulation

Within days of its public release, one of the most advanced AI models ever built was forcibly pulled from the market. Not by its creator, but by the United States government. The model in question was Fable 5, the public-facing version of Anthropic's frontier system known as Mythos and its sudden disappearance set off a chain reaction that has exposed just how fragile, inconsistent, and unprepared America's approach to AI regulation really is.

A Model Yanked, A Company Blindsided

Anthropic, the AI safety company founded by former OpenAI researchers and valued at nearly a trillion dollars, had spent months preparing for the release of Mythos, a model the company itself acknowledged was extraordinarily capable at identifying cybersecurity vulnerabilities. Understanding the sensitivity of the technology, Anthropic released the full version of Mythos only to a small group of vetted partners. For the broader public, it released a restricted version called Fable 5 on June 9, complete with additional guardrails designed to prevent misuse.

The Trump administration, however, was not satisfied. After being notified of a jailbreak a technique that allows users to bypass a model's built-in safety constraints the administration declared Mythos a national security risk and moved swiftly. Anthropic, according to a source close to the company, was initially given just 90 minutes to comply. Within days, both Mythos and Fable 5 were pulled entirely from customers.

What made the situation even more striking was that the government provided no specific details about the nature of the national security threat. Anthropic, operating in the dark, had little choice but to comply.

Anthropic Pushes Back

Anthropic did not go quietly. In a public statement on June 12, the company argued that the vulnerabilities identified by the government were minor in nature and not unique to its models -- meaning similar weaknesses exist across products from other AI companies as well. The implication was clear: if the standard being applied to Fable 5 were applied universally, a significant portion of the AI industry could face the same fate.

The company's position found support in unexpected places. Dozens of cybersecurity researchers, AI entrepreneurs, and corporate executives signed an open letter posted at freefable.org, sharply criticizing the government's actions. The signatories did not dispute that the government has a legitimate role in assessing AI safety risks. What they objected to was the process, or rather the complete absence of one.

"To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous," the letter read.

Alex Stamos, the former chief security officer at Facebook and a respected voice in cybersecurity circles, went further. After reviewing the research that informed the government's decision, Stamos posted on X that while the findings had some validity, there were "no unique capabilities that justify a reaction close to this."

The Amazon Angle

Adding another layer of complexity to the story is the question of how the jailbreak was first flagged to the government. A source familiar with the situation told CNN it was Amazon that initially brought the vulnerability to the government's attention. Amazon is a major investor in Anthropic and also a direct competitor in the AI space through its own AWS AI products. Whether Amazon's notification was routine responsible disclosure or something more complicated remains unclear, but it has prompted questions about the relationship between commercial interests and national security processes.

Trump Weighs In

Despite the administration's initial alarm, the situation appears to be de-escalating. Anthropic and the government have been in ongoing talks, and at the G7 summit, President Trump said negotiations were "going fine." In an interview with Axios published on June 20, Trump went further, saying he no longer views Anthropic as a national security threat. "Well, not now, but a week ago, maybe," he said.

Trump adviser David Sacks, who previously served as the White House AI policy czar, has taken a harder line. Sacks pushed back on Anthropic's characterization of the jailbreak as minor. "It's difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not 'serious,'" he wrote on X.

The White House referred CNN to the Commerce Department for comment. The Commerce Department did not respond.

A Regulatory System Without a Spine

The Anthropic episode is not an isolated incident. It is the latest in a series of confrontations between the company and the federal government that paint a picture of a regulatory relationship built on improvisation rather than principle.

Earlier this year, Anthropic found itself in a dispute with the Pentagon over the military's request to modify the guardrails in Anthropic's AI systems for defense applications. When Anthropic declined, the Department of Defense responded by labeling the company a "supply chain risk" and placing it on a blacklist. That conflict has never been fully resolved, and it set the stage for the current standoff.

Zooming out, the broader regulatory landscape is a patchwork of competing and often contradictory approaches. The Trump administration has leaned heavily toward deregulation, rolling back Biden-era requirements for mandatory safety reporting in favor of voluntary frameworks. In March, the administration issued a national policy blueprint for AI that explicitly recommended against creating a single AI regulatory body, instead calling for sector-specific oversight a structure that, as the Anthropic situation demonstrates, can leave enormous gaps.

An executive order issued earlier in June asked AI companies to voluntarily share advanced models with the government for cybersecurity review before public release. Even that modest step was delayed at the last minute after Trump expressed concern it might slow down American AI development.

Meanwhile, states have taken matters into their own hands. California passed legislation requiring AI companies to publish risk frameworks, report safety incidents, and protect employees who raise concerns. Florida, in a more dramatic move, opened a criminal investigation into OpenAI and filed a civil lawsuit alleging that ChatGPT may have played a role in the mass shooting at Florida State University. OpenAI has denied the allegations. Whether those state-level efforts represent genuine regulatory progress or political grandstanding is itself a matter of debate.

The Cost of Opacity

What connects all of these episodes is the absence of a transparent, consistent framework for making decisions. Brad Carson, who leads Public First, a bipartisan pro-AI safety political action committee, described the current situation bluntly: "Right now, you have an ad hoc, personalized, opaque, possibly lawless approach."

The most pointed critique has come from legal scholar Jessica Tillipman, associate dean for government procurement law at George Washington University. In a widely read essay, Tillipman acknowledged that national security decisions require government discretion. But she argued that discretion without process is not governance it is something more dangerous.

"The problem is not that the government exercised discretion; national security demands such latitude," Tillipman wrote. "What is striking is the absence of any meaningful process."

She went further, warning that the administration's handling of Anthropic could trigger the kind of regulatory backlash it is trying to avoid. "The damage does not stop at one firm. An administration that governs this way will not avoid the heavy regulation it fears. It is manufacturing the conditions for catastrophe or abuse that, in every cycle I've documented, triggers exactly that response."

What Comes Next

The immediate crisis appears to be easing. Talks between Anthropic and the administration are ongoing, and Trump's softened rhetoric suggests a resolution may be in sight. But the underlying structural problems have not been addressed. There is still no clear process for how the government evaluates AI safety risks. There is still no consistent standard for when a vulnerability justifies pulling a product from the market. And there is still no framework that gives companies or the public a reliable way to understand what rules they are operating under.

For Anthropic, which is widely expected to pursue an initial public offering in the near future, the stakes extend well beyond this particular dispute. A company worth nearly a trillion dollars, operating in a regulatory environment defined by unpredictability, faces a uniquely difficult challenge: how to build trustworthy AI when the rules can change in 90 minutes.