It’s no secret that large language models (LLMs) and generative AI have become a massive and critical part of the American and global economy, with the St. Louis Fed reporting that AI accounted for 39 percent of GDP growth in the first three quarters 2025. It has also become a hot-button issue in politics, often at the local level, with debates raging about data centers in town hall meetings across the country. However, the US government’s recent directive to ban foreigners from accessing the latest Anthropic models makes obvious what some already understood: that AI development itself has become deeply political. What many may not know however, is that this is part of a long-running marketing effort by America’s leading AI labs, which is now beginning to backfire.
On April 7, 2026, Anthropic—one of the so-called “frontier labs” competing with OpenAI and Google—announced that they had developed a new model named Claude Mythos. However, there were no plans at the time to release Mythos to the public, given the model’s vastly-improved performance in discovering and writing exploits for security vulnerabilities.
Instead, the company announced Project Glasswing—a limited pilot which allowed the model to be accessed by a selected group of big tech corporations and, eventually, the US Government. The announcement was a major marketing coup, with the panic over cybersecurity giving Mythos an alluring mystique, and reinforcing the idea that Anthropic made the most capable models in existence. However, something more sinister was at play. By portraying AI as inherently dangerous, but themselves as uniquely responsible, Anthropic was inviting the government to tip the scales of the market in their favor.
The Mythos announcement was not the first time frontier labs marketed AI by portraying it as dangerous. In 2019, before the recent AI revolution in which LLMs became practically useful, OpenAI publicly announced that they would not publish GPT-2 over fears of misuse. Later, in 2023, leaders from all the frontier labs released a joint statement describing AI as an extinction level threat comparable to nuclear war. This hysterical messaging often came with the call for additional regulation.
While testifying before congress in 2023, Sam Altman—the CEO of OpenAI—suggested that the government impose licensing and testing requirements for AI models above a certain capability. Anthropic endorsed a similar idea in their response to a request for comment from the National Telecommunications and Information Administration.
In the aftermath of Project Glasswing, however, Anthropic took the calls for regulation to a new level. They revised their “Policy on the AI Exponential” to call for stricter regulation, published an article calling on the US government to do more to stop the rise of Chinese AI while playing into tired war propaganda, and even called for a government-mediated pause on frontier AI development (while they of course are in the lead).
It is worth noting that this behavior is not necessarily completely cynical. Many crucial figures in AI development—most notably Anthropic CEO Dario Amodei—truly believe that AI may pose an existential threat to human civilization, and that the safest option is for western governments to take action that keep responsible decelerationists (such as Dario himself) in control of the market. Regardless of motive however, Anthropic are advocating for horribly-authoritarian policies which stand to make them a lot of money.
Libertarians and classical liberals are very familiar with such tactics. To quote Rothbard,
We often fail to realize that the point of much of Big Government is precisely to set up such “partnerships,” for the benefit of both government and business, or rather, of certain business firms and groups that happen to be in political favor.
So-called public-private partnerships of this nature were a defining aspect of the Progressive Era which gave us the incredibly powerful federal government that is so dominant over our lives today, and thus played a vital role in the history of the United States and its transformation from a uniquely-free country into a modern empire. Indeed, the advocates of laissez-faire have long warned of the authoritarian society that would come from such relationships, destroying even the business class which advocated for it. As Frank Chodorov wrote,
What must be the end-result of the mesalliance of business and politics? Just as bad money destroys the value of good money, so the virus of political intrusion into the body economic must undermine its health and ultimately wreck it. Private enterprise must go.
Chodorov’s warning rings true. As the greatest champions of AI regulation, Anthropic has been the company most targeted by the current regime. Back in March, Anthropic lost out on a major Pentagon contract and was labeled a “supply chain risk” for refusing to agree to a government contract which allowed the government to use their models in autonomous weapons systems.
Notably, it seems that the Mythos marketing blitz was in part an attempt to regain favor with the government, as the promise the model showed for cybersecurity purposes proved too much for the NSA to turn down despite the recent designation. The gambit apparently failed, however, when Antropic released their Fable 5 model to the public—a heavily-guardrailed version of Mythos designed to refuse all requests related to cybersecurity or biology. On June 12, following reports to the government from Amazon that the model could be jailbroken, the US government issued a directive banning all foreign nationals, including Anthropic employees, from accessing Fable or Mythos.
In the days following the Mythos export ban, the situation has evolved in a very concerning way. Anthropic released a statement reacting to the state directive in which they seemingly walked back their claims of the model being some sort of cyber superweapon, comparing the model’s cyber capabilities to OpenAI’s flagship model GPT 5.5. They also state their belief that the ban was triggered by the existence of a jailbreak for Fable’s guardrails. This point is important because such a jailbreak almost certainly takes the form of prompt injection which is not a solvable problem, and will always represent a gap in AI security.
Given this, it does appear possible that all future advancements in American-made AI, no matter how tightly-guardrailed, will be heavily restricted by the federal government. Or even that such restrictions could expand to include other highly-capable models which are currently available. Fear of this outcome was heightened when Anthropic updated their privacy policy to include language around ID verification, which will be a necessary step if they wish to make Fable 5 available to US citizens while export restrictions are still in place. Overall, the situation appears to have led to a circumstance where the federal government has far more control over the AI market than they have exerted in the past.
Anthropic isn’t the first company to try these tactics and they won’t be the last. As mentioned above, other AI labs have engaged in fear-based marketing with an eye towards regulation. Big business colluding with big government to secure exclusive privilege is nothing new. It is, however, a threat to the liberty of every American. By empowering the state to destroy their competitors, big business creates a monster beyond their control.
Intervention begets more intervention. Unfortunately, the capabilities of the latest AI models are very real, and very impressive. While Mythos may not be the superweapon Anthropic claimed, it is by all accounts a very powerful tool. The government’s latest actions have made AI a more important battleground than ever, as those out of favor with the state may soon need to scramble to secure access to the technology before it is too late.
