This op-ed was originally published in Real Clear Markets.
The work of writing an op-ed on regulatory matters usually proceeds thus: officials decide on a policy; the analyst reads the policy (one hopes), weighs its merits and demerits, and formulates an argument; finally, that argument appears in a newspaper or digital publication to inform the public.
The Trump administration’s new framework to inspect advanced artificial intelligence (AI) models before they are released to the public has rendered several of these steps impossible. The White House, it seems, has declined to make the framework public, precluding most analysis of its particulars. What little the public knows at present about this document can be collected from the fragmentary and often vague comments unnamed sources have provided to the media. So few and disparate are the tiles seen by the public eye that the full mosaic image remains unidentifiable. The regulation of the vanguard of the most significant technology of the age will be undertaken without public knowledge of its standards or, likely, any chance for the public to assess or comment on the standards’ implementation.
A few observations can be made on the basis of what is known, few of them complementary to the framework’s character. But no full analysis may be given, for there is little to analyze. That, in itself, should be deemed a damning defect.
It is doubtful that AI firms seeking to market advanced models will know of the obligations imposed by the framework. “[D]etails will only be made available to the companies that are part of the process,” Axios reports, and itis reported that the White House has briefed some firms. And yet, if the framework remains the architecture of pre-deployment review, an insurgent in the AI industry, otherwise prepared to challenge incumbents, would be relegated to ignorance until the fatal moment at which they elect to become “part of the process.” These insurgents will likely be constrained, playing the odds and not analyzing known facts, to enter a dark room in which regulators lurk, the requirements (and perhaps the outcome) of the review process which awaits them unknown.
Such uncertainty is anathema to innovation—to dynamism and capital investment generally, in fact. To hide the regulatory costs that accompany innovation is to dissipate the entrepreneurial energies that sustain America’s present technological dominance, to dam up the headwaters of innovation itself.
Still more uncertainty likely awaits American innovators, incumbents and upstarts alike. Even knowing the review standards and procedural requirements of today, regulated firms cannot anticipate those of tomorrow; the corollaries of secretiveness in regulatory endeavors quite often are plasticity of policy and arbitrary enforcement. Sure enough, the White House did not wait to implement the framework before hinting at future changes: “A.I. models, which have computer code available to the public to download and modify [i.e., open source models], although that could change as the technology advances,” sources told The New York Times.
Worse still, the administration, whether by incompetence or by design, seems to have allowed itself broad latitude, iteratively and continually, to tailor enforcement of the framework to satisfy impulses, no matter how momentary. To wit, the White House reportedly indulged in rampant vagueness, always dangerous to regulatory clarity and certainty. For example, notes Axios “[t]here is no clear definition of what is considered state-of-the-art or a national security risk.” And the confusion has gained the attention of industry: “The general definition of state-of-the-art capabilities could be interpreted differently by makers of closed models and the White House, potentially creating confusion about which models fall into the category,” The Wall Street Journal (WSJ) was informed. Moreover, it is “unclear which ‘trusted partners’ will get early access to advanced models under the framework, including whether any foreign governments would qualify.”
For innovators, there will be no avoiding these difficulties. The framework is to be voluntary, it has been reported. But as an advisory government is a contradiction in terms, voluntary standards often prove to be anything but. “While the testing is voluntary, industry executives have said it would be risky to bypass the government review, particularly if there were safety concerns raised after the fact,” WSJ reports. The Trump administration, acting unilaterally, already has hazarded an attempt to freeze Anthropic’s Fable 5 and Mythos 5 before release; the anxieties of industry executives are not unfounded.
Thus lies the field: the Trump administration has determined, without Congress, to arrogate to itself the authority to “review” the best and newest of the AI industry. It will not inform the public how the task will be accomplished, nor does it seem likely to leave its framework untampered-with. It will enforce the reportedly vague language it drafted as it sees fit—the only judge of its performance being itself. All the while, AI firms will find no respite; escape from this Kafkaesque regulatory structure invites economic extinction. “Voluntary” compliance will be enforced.
“We don’t want to restrict them where all of a sudden we come in second to China,” President Trump said recently, speaking of American innovators. These aspirations notwithstanding, his framework may prove the accomplishment that wins America the silver medal.
David B. McGarry is the research director at the Taxpayers Protection Alliance.