DOGE is Overpromising on Deregulation

In a recent podcast on DOGE 1.0, Alex Nowrasteh and I discussed how Elon Musk had overpromised on government spending cuts and so inevitably under-delivered.

Given the composition of the federal budget, and without a legislative strategy to work with Congress to cut entitlement programs and defense spending, it was simply impossible for DOGE to ever eke out $1-2 trillion of financial savings from cancelling DEI grants, laying off federal employees, shuttering minor departments and agencies, and delivering various software and anti-fraud upgrades to federal services.

Indeed, making such wild promises on spending arguably undermined focus on what it did help achieve: a significant reduction in federal headcount, scrapping wasteful economic development aid, and zeroing near-term public broadcasting subsidies (since codified in the rescissions package).

Would DOGE 2.0 heed the lesson not to overpromise again? It appears not.

The Washington Post has unearthed a DOGE PowerPoint dated July 1 called "DOGE Deregulation Opportunity." It contains what looks like a logically flawed estimate suggesting that a DOGE deregulation drive could deliver a mammoth $3.3 trillion per year in economic benefits.

The story goes like this. The DOGE team has built an AI model which can parse existing legislation and the federal regulatory rulebook to identify regulations that are a) not required by statute and b) not required for the effective operation of departmental activity. By using the AI tool to identify them, policymakers can then work through the list and go about simply deleting them.

No doubt, there’s promise here. State and local governments in Ohio, Virginia, and San Francisco are already using AI to prune outdated rules. One obsolete section of San Francisco’s municipal code required biennial reports on the state of the city’s newspaper racks – which haven’t existed since February.

So I am in little doubt that, done well, AI could help identify some superfluous regulation and clean up the regulatory code [“done well” is an important qualifier: Cato’s AI experts tell me existing AI tools still often struggle to maintain context with large volumes of input like this, especially with overlapping and interdependent regulations].

But would such efforts generate anything close to $3.3 trillion per year in economic benefits? Color me skeptical.

DOGE’s Calculations

DOGE draws on external sources to assume that regulation costs America annually:

· $3.1 trillion in compliance costs,

· $1.2 trillion in reduced investment,

· $2.2 trillion in lower sales revenue, and

· $175 billion in federal enforcement costs.

This, of course, ignores offsetting benefits, but that’s beside the point.

Next, the agency proffers that 50 percent of all federal regulations are neither required by statute nor needed by the agencies enforcing them. With this, DOGE engages in rudimentary arithmetic for its topline number. It sums the overregulation costs to just over $6.5 trillion and divides by two—since half of regulations are unneeded, duh—for its $3.3 trillion per year figure in benefits. In other words, per this calculation, we can grow the level of GDP by more than 10 percent with an AI deregulatory quick fix!

Not All Regulations Are Created Equal

Now, this methodology is self-evidently flawed. First: regulations not required by statute and which are genuinely unnecessary for departments to administer are unlikely to be those that impose high economic costs.

Consider one quick case study from banking law. The Federal Deposit Insurance Corporation reports that through 2024, 47 percent of “Regulation E” violations cited a single section of the regulation (establishing timelines for banks to investigate disputed card charges). DOGE could shred much of this regulation, but if it keeps just this section, comprising only 1.5 percent of the regulation’s total word count, the supposed clean-up would do little.

DOGE could, of course, improve its efficiency by targeting operative super-sections like this, but the exact opposite is more likely. Remember, DOGE is promising to prioritize regulations not required by statute for repeal, but this example is directly on-point as far as statute goes. Congress’s express purpose was to cement a bank’s responsibilities in transactions using the new payment card technology (the act passed in 1978).

Presuming, as DOGE has, that cutting half the text of the Code of Federal Regulations will halve the costs of regulation is therefore a bit like thinking removing half the cars from parking lots and driveways throughout a city will halve traffic congestion. It will no doubt help directionally, but not all cars contribute to congestion equivalently.

Inflated Compliance Cost Estimates

Second, DOGE’s claim of $6.5 trillion+ in yearly regulation costs is itself bloated with double counting.

Take their “compliance costs” estimate. It’s really nothing of the sort. It leans on a National Association of Manufacturers (NAM) study which estimates the total GDP losses from economic regulation in an indirect way.

NAM bases its estimate on a shaky Regulatory Index from the World Competitiveness Center, in turn based on fuzzy business executive surveys about regulation. At bottom, NAM uses regression analysis to estimate that the U.S. suffers from $2 trillion in lower GDP from regulation compared to being the country in the world where execs are most satisfied with their regulatory regime.

In other words, DOGE treats a speculative estimate of GDP losses as a hard cost on businesses. In fact, it chalks up this GDP loss estimate as economic regulation’s “compliance cost,” before piling on additional hits for slashed investment, weaker sales, and bloated government spending.

The problem is, investment, sales, and government spending are baked into the GDP loss estimates already, so DOGE is engaged in a large dose of double-counting. Yes, GDP might miss some regulatory distortions (and indeed regulatory benefits), but I think it’s safe to assume DOGE is hugely overestimating here.

Deregulation is still an admirable goal

Deregulation is nevertheless an admirable and achievable goal for the Trump administration. The total cost to the U.S. economy of regulatory compliance and deadweight losses may very well total trillions of dollars, but is unlikely $6.5 trillion. And there are already some bright spots already for deregulation too. The attempt to rescind the EPA’s endangerment finding could be significant, and the appointment of Casey Mulligan, an economic expert, to the Small Business Association, is a signal that Trump’s White House is taking business deregulation seriously.

Yet DOGE would be wise not to overpromise what its AI systems can achieve. Yes, we need regular efforts to cull defunct rules and regulations from the code, and a successful effort here will reduce compliance costs and other economic distortions. But it’s not mainly the obsolete or superfluous sections of the regulatory code dragging on economic activity; it’s the active sections!

Just as making deep spending cuts always required cutting real programs rather than just taking aim at waste, fraud, and inefficiency, so too will a major deregulation drive require Congress legislating for fewer federal goals, reclaiming Article I powers from agencies, and engaging in the hard task of rewriting rules while cognizant of their interdependence.

AI certainly has a role to play in all this, but by pretending there are trillions left on the table by not simply expunging unnecessary rules, DOGE risks making overpromising and under-delivering a habit.