Stop rogue government AI

Performative Washington, responsible for funding AI weaponry and R&D for years, purports to have suddenly discovered the dangers of artificial intelligence.

Recent incidents have raised legitimate cybersecurity concerns on an ill-equipped public Internet, an Internet whose architecture was substantially shaped by government-backed development and helped produce today’s universal, centralized network rather than more decentralized private “Splinternets,” with greater scope for exit, differentiation, and resilience. That may have left us more exposed to some of the AI risks now emerging.

But now some of the leading voices in AI, including companies doing substantial business with the federal government, it must be said, are warning about risks and calling for yet another layer of government over Washington’s already significant existing role in steering AI. This seems too convenient in some ways.

The question Washington should be asking is not simply, “How do we regulate AI?” It is: “Who is already funding, deploying, purchasing, directing, and regulating AI, and what happens when all those functions fully converge within government?”

Official Washington’s blatant coalescing around an expanded government role in AI this week cannot be called naïve. People involved know full well the machinery they are building atop an already extensive architecture of business-government entanglement, one in many ways detached from competitive free enterprise.

The government is hardly a disinterested outside observer of these risks. Washington has spent decades funding, procuring, and deploying advanced technologies for military and intelligence purposes. Vocal politicians are not merely confronting AI as detached would-be regulators; they represent a government that is itself a funder, developer, and customer, in ways that compromise them as would-be rulemakers. That makes the newly energized rush to build another layer of AI governance more consequential than it may appear at first glance. Our leaders have already helped accelerate a technology in ways that can insulate it from market disciplines yet now invoke the risks surrounding it as justification for expanding their own culpable governing machinery as a remedy.

Reps. Mike Lawler (R-NY) and Josh Gottheimer (D-NJ) recently introduced the bipartisan “Stop Rogue AI Act” on September 14 (here’s their September 17 press conference), which would, to its credit, not institute a pause or “kill switch.” However, even as the administration otherwise pursues shrinkage of the federal bureaucracy, such legislation would give new regulatory oversight to the National Institute of Standards and Technology (NIST): directing NIST to develop national standards, guidelines, and best practices for discovering, verifying, monitoring, and controlling AI agents. According to Lawler’s press release, federal agencies and contractors would then “incorporate those safeguards into how they procure and deploy AI,” with NIST and the Cybersecurity and Infrastructure Security Agency folding the standards into existing federal cybersecurity guidance.

The sponsors describe this as putting “humans back in the driver’s seat.”

Granted, lawmakers and bureaucrats are people too. But this may be how an intractable AI regulatory apparatus gets built: first standards, guidance, and procurement requirements, then expectations that could migrate outward into the broader private sector and startup community. This new legislative proposal is only one of many, and the widespread but misdirected call to “stop rogue AI” can easily evolve into a top-down institutional infrastructure for deciding what AI is permitted to do, who may deploy it, under what conditions, and according to whose definition of safety.

In other words, when government becomes AI’s funder, customer, standards-setter, infrastructure partner, and potential guarantor or indemnifier, we get misalignment by design rather than regulation: the deliberate fusion of government power, corporate interests, and social management that removes AI from true market discipline rather than subjecting it to normal controls. This is the AI Industrial Complex toward which we are hurtling, “rogue” in its own right.

If policymakers think they are going to construct the equivalent of an AI bill of rights, they should first be capable of articulating why their primary job is to constrain government’s appetites over it, rather than presume that private competitive enterprise is the problem to be managed. Government’s own AI funding, procurement, surveillance, military and intelligence applications, and deployment deserve most of the scrutiny at this juncture.

The irony is hard to miss. Policymakers profess to be frightened that AI agents might become too autonomous, so their answer is to create an increasingly autonomous, unelected government AI apparatus to monitor them, shaping conduct through layers of guidance and procurement rules without necessarily passing through ordinary legislative or regulatory notice-and-comment processes.

The proper response to rogue AI is therefore to sharply constrain government’s own use, funding, procurement, surveillance, and deployment of AI, and to prevent a new governance architecture from becoming entrenched in the first place. That’s draft legislation we’ve yet to see.

For private misconduct, we already have well-established rule-of-law institutions if we choose to use them. Property rights, contracts, fraud, trespass, negligence, cybersecurity law, criminal law, liability, and surely other categories beyond what I can name all remain available to address actual harms.

If AI alignment matters, then start by aligning policies with limited government and the Constitution; concerns heretofore all but unmentioned in the AI debate. Otherwise, the cure for rogue AI may be nothing of the sort. It will instead be the birth of something considerably more entrenched: the AI administrative state.

For more:

“Misalignment by design: How AI could become the operating system of the administrative state,” (Part 1) (Part 2)

“Artificial Intelligence Model Legislation and Bill of Rights Regulating Government—Not Private Competitive Enterprise“ Social Science Research Network

“Welcome to the Machine: How Schemes to Control AI Are Progressive Tools to Control Society,” Competitive Enterprise Institute

“Trump’s AI plan clears the field—then occupies it,” Competitive Enterprise Institute

###