The drumbeat for regulating artificial intelligence is growing louder by the day, largely due to a coordinated campaign that magnifies the technology’s risks. It ignores the critical question: Can we regulate effectively without causing a disaster for our technology leadership and economy?

That is not to say, however, that some form of control to prevent the worst consequences of AI — especially “superintelligent AI” — is not needed. More on that below.

Four serious problems arise from creating any form of AI regulatory entity.

The first is “regulatory capture.” Except for employees in the leading AI companies and a few academics, the expertise to staff this body is scarce. Any mechanism for review, pre-clearance, or rule enforcement will thus be beholden to the targets of regulation — the AI companies. The fox will be renting luxury accommodations in the hen house.

Second, the current ambiguous proposals for a regulatory framework are questionable in their effectiveness. Existing regulatory regimes, whether for energy, health, construction, etc., often involve a tortuous process, sometimes with contradictions.

Typical regulatory incentives aim to avoid harm even at the cost of progress, and to minimize political pushback. Regulators reflexively regulate as much as possible. (As an FDA commissioner once said, “Cows moo, dogs bark, and regulators regulate.”) That is a perfect prescription for losing U.S. AI leadership in a critical field moving at hyper speed.

Third, regulations rarely keep up with progress. The 1996 Communications Act sheltered a nascent online publishing industry from any accountability for content, but did not anticipate how social networks and search providers would burgeon, make their products addictive, and fail to recognize the consequences. That has enabled self-serving influencers, grifters and exploiters of children, and has created vast numbers of troubled and alienated young people.

Fourth, delegation of power to resolve regulatory ambiguities can run amok and become myopically focused. This shows up in political tinkering with broadcast licenses, bizarre interpretations of what constitutes a navigable waterway, and innumerable other dubious actions that fly under the radar but have significant real-world consequences. New York City has tens of thousands of vacant apartments where improvements are economically impossible under current building codes and conflicting constraints of rent control. Such regulation serves nobody.

AI regulation is a minefield that could lead to unintended consequences: undermining some AI vendors, incentivizing a switch to open-source models that would escape controls or guardrails, distorting the economics of various parts of the AI ecosystem, etc.

However, there is a need to deter the misuse of AI — especially of “superintelligent” AI — to hack protected systems, create weapons, swindle or blackmail innocent parties or companies, and more. The approach to regulation that would accomplish that is carefully crafted incentives and disincentives with meaningful consequences.

The only continuously adaptable mechanism we have to meet such challenges is the legal system — including civil and criminal law — if it is built with the right framework. We need to avoid re-creating the worst aspects of law: ambiguous infractions, frivolous lawsuits, the predatory tort bar and regulation by litigation.

On the criminal side, there should be severe penalties for misusing AI to produce illegal or dangerous products, harm others or misappropriate money or property. AI is a powerful tool, and its misuse must be punished accordingly, with few escape hatches.

Civil law should define malpractice as it applies to AI developers. Legal culpability should not apply to AI that simply produces inaccurate responses to prompts. The focus should be on negligent design and testing that creates entire classes of problems. Designating those with standing to sue is a topic for careful consideration.

We also suggest creating a tier of well-compensated special federal magistrates to screen all civil actions related to AI. The AI companies should defray the cost of training these new recruits.

Trusting Congress to craft complex and carefully constructed legislation may be wishful thinking. Perhaps the AI companies would assist out of goodwill even though no competitive benefit would accrue. They claim to care, after all.

We make no prediction about how likely this approach is to be adopted or how well it would function. We do believe that creating any form of regulatory entity out of panic would be a historic governance mistake.

_____

ABOUT THE WRITERS

Andrew I. Fillat is the co-inventor of relational databases. He wrote this for InsideSources.com.

Henry I. Miller is the Glenn Swogger Distinguished Scholar at the Science Literacy Project. He wrote this for InsideSources.com.

_____

©2026 Tribune Content Agency, LLC