A Letter from Rhino Times reader Thomas Knapp, the director and senior news analyst at the William Lloyd Garrison Center for Libertarian Advocacy Journalism
“The US government,” artificial intelligence firm Anthropic informed the public in a June 12 statement, “citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. “
As of June 15, according to Just Security, the government isn’t allowing the public to see what’s actually in that directive, but according to Anthropic, it cites concerns that the company’s models are vulnerable to “jailbreaking” that would let users get around “guardrails” that prevent them from answering certain kinds of questions (obvious example: How to successfully execute a terrorist attack).
Whatever the real reasons for the directive — the move looks, on its face, less like a real “national security concern” and more a revenge move against Anthropic for refusing to let the Pentagon use its models in autonomous weapon and mass surveillance projects — it’s both a bad idea and an unambiguous violation of the US Constitution’s First Amendment’s free speech protections.
A syllogism:
Code is speech (as ruled by a US district court and affirmed by the Ninth Circuit Court of Appeals in Bernstein v. Department of Justice).
AI models are code.
Therefore, AI models are speech, and the government doesn’t get to control them.
Not that the current administration, or any other, or Congress, or the courts, can be counted on to respect the Constitution. The ink wasn’t dry on that document before the American political establishment started ignoring its inconveniences.
Which leaves Anthropic and other artificial intelligence firms in a bind. At every point in their development of better models, they’ve had busybodies and bureaucrats peering over their shoulders, nudging them in various directions and cuffing their hands when the nudges don’t work.
As a legal matter, I describe the problem above.
As a practical matter, if Anthropic et al. want to innovate and compete in a growing market that’s already changing how the world works, they need to get away from the US government, which means getting away from the US.
They should re-domicile their companies to, and move those companies’ operations to, places beyond the long reach of Uncle Sam.
Money may not buy happiness, but in certain contexts it can probably buy substantial freedom. There’s lots of money in AI. There’s going to be more.
It’s a big planet, and while much of it groans beneath the rule of authoritarian regimes like the US, the People’s Republic of China, and the Russian Federation (among others), there’s almost certainly a government SOMEWHERE possessed of the common sense to accept golden eggs without strangling the geese that lay them.
These firms should look for governments willing to offer non-interference pledges in return for infrastructure investment and a reasonable tax rate.
One long-term alternative is moving AI infrastructure not just offshore, but off-planet, mostly beyond the control of ANY government, but we may be decades away from that as a practical option.
Remember: If something can be done, it will be done. If it’s not done by one of the large US AI firms, it will be done somewhere else and/or by someone else, to the detriment of those firms and quite possibly to the detriment of their American customers.
My own concern is less with the future of Anthropic, OpenAI, et al. than with the US regime’s perpetual attacks on speech in general and on code AS speech. My first experience with the latter came during the regime’s attempts to “contain” strong encryption with export controls in the 1990s. Freedom fighters beat them then, and can beat them now.
Thomas L. Knapp (thegarrisoncenter.org).
However, that concern does not automatically mean AI models should be freely available to anyone, anywhere in the world.
Unlike ordinary speech, advanced AI systems can provide capabilities that have direct military, intelligence, cyber-security, and weapons-related applications. Nations have long restricted exports of technologies with strategic value, including cryptography, advanced semiconductors, missile technology, and nuclear-related research. Whether we like those restrictions or not, they are based on the idea that some technologies are more than mere expression…they are also tools.
Knapp’s First Amendment argument is also weaker than he suggests. It is true that courts have recognized computer code as a form of speech. However, courts have never treated all code exactly the same as political speech. The government has historically been allowed to regulate the export of technologies that have both expressive and functional characteristics. The fact that code can be speech does not automatically mean it cannot be regulated when it functions as a strategic technology.
The Bernstein case involved encryption software and established that code contains expressive elements. It did not create a blanket rule that any software or AI model must be freely exportable worldwide. If it did, export controls on encryption, military software, cyber tools, and countless other technologies would likely have been struck down decades ago.
The stronger argument against the administration is not that AI can never be regulated. It is that any restrictions should be transparent, narrowly tailored, based on demonstrable security risks, and applied consistently rather than selectively. Governments should have to show why a particular AI capability presents a genuine national security concern instead of simply asserting that it does.
In short, I share Knapp’s skepticism toward government secrecy and potential political retaliation. But I do not share the view that advanced AI should be available without restriction across international borders or that the First Amendment creates an absolute right to export any AI system to anyone in the world.