Gas-Powered Leaf Blowers and the Real Lesson for AI Rules

Every fall, the same sound returns to American neighborhoods: a two-stroke engine, unmuffled by any catalytic converter, running somewhere between 95 and 105 decibels at the operator's ear. Most of us have learned to tune it out. But a recent essay in The Regulatory Review, by law professor Mark Nevitt and titled "The Case Against Gas-Powered Leaf Blowers Has Only Grown Stronger," makes an argument worth sitting with. The evidence against these machines hasn't just persisted. It has accumulated, and accumulation is a different thing than a single bad finding. It's the difference between a claim and a pattern.

I want to use this small, almost comic policy fight — grown adults strapping two-stroke engines to their backs to move leaves from one side of a lawn to the other — to say something bigger about how institutions respond to mounting evidence generally. Because the leaf blower case isn't really about leaf blowers. It's about the gap between what we know and what we regulate, and that gap is exactly the terrain AI governance is standing on right now. The mechanism, the industry pushback, and the slow institutional catch-up all look familiar if you've been watching how AI rules have arrived over the past two years.

What the Evidence Against Gas Leaf Blowers Actually Says

Start with the plain facts, because they're more damning than the framing usually lets on.

Gas-powered leaf blowers, like most handheld lawn equipment, run on two-stroke engines. Unlike a car engine, a two-stroke mixes oil directly into the fuel and burns both together. That's mechanically simple and cheap to manufacture, but chemically filthy: a two-stroke engine has no dedicated exhaust stroke, so a portion of the unburned fuel-oil mixture escapes straight out the exhaust port. The California Air Resources Board has made this point in stark terms. Operating a commercial gas leaf blower for one hour produces smog-forming pollution comparable to driving a 2017 Toyota Camry about 1,100 miles. That comparison comes from CARB's own testimony and public materials supporting its small off-road engine rulemaking, not from an advocacy group's rounding-up.

Noise is the second half of the case, and it's the half most homeowners actually experience. The Occupational Safety and Health Administration sets a permissible noise exposure limit of 90 decibels averaged over an eight-hour workday, under 29 CFR 1910.95. Manufacturer spec sheets for commercial backpack blowers routinely list operator-ear noise levels in the high 90s to low 100s. For a landscaping crew running several units side by side for hours, that threshold isn't a distant hypothetical. It's a daily condition of employment.

None of this is new information, and that's the actual point Nevitt is making, and mine. Engineers have understood the emissions profile of a two-stroke engine for decades. What's changed is the accumulation of secondary evidence, particularly around the labor exposure landscaping crews absorb every shift. The evidence didn't get discovered. It got harder to ignore.

Two Ways to Regulate the Same Harm

California and Washington, D.C. reached similar conclusions about gas leaf blowers, but they chose different regulatory tools, and the difference matters if you want to understand how any government translates accumulated evidence into a rule.

California's Assembly Bill 1346, signed in 2021, directed CARB to prohibit the sale of new gas-powered small off-road engines, a category that includes leaf blowers, mowers, trimmers, and generators under 25 horsepower. The phase-out took effect for engines manufactured on or after January 1, 2024. The law doesn't ban equipment you already own or make it illegal to use a gas blower today. It closes the spigot on new sales, which is a slower, more politically durable form of change than an outright ban.

Washington, D.C. took the more direct route. Its Leaf Blower Regulation Amendment Act, passed in 2018, banned the sale and use of gasoline-powered leaf blowers within the District effective January 1, 2022. That's a use ban, not a sales phase-out, and most local ordinances that followed have borrowed from it, usually scaled down to seasonal restrictions or noise thresholds rather than a full ban.

The federal layer beneath both is thin. The Environmental Protection Agency regulates small nonroad spark-ignition engines under 40 CFR Part 1054, but those standards were built around the two-stroke architecture rather than around phasing it out. That's exactly why the more aggressive moves have come from states and cities, not Washington. I keep running into this same shape across very different regulatory domains: the federal floor moves last, and when it moves, it often just formalizes what the technology already does rather than pushing it somewhere new.

Why Known Harms Take So Long to Become Regulated Harms

Here's the part worth sitting with. The mechanism behind two-stroke engine pollution was never in dispute. Nobody credible argues that these engines are clean or quiet. What actually drove the roughly decade-long gap between the first serious state-level proposals and California's 2021 law wasn't a scientific question. It was a distribution question: who pays for the switch, and on what timeline.

The National Association of Landscape Professionals opposed AB 1346, citing equipment cost and battery runtime limitations for commercial-scale crews. Those are real operational concerns, not manufactured ones. But it's worth naming the distinction plainly. "This transition will cost us money and time" is a legitimate argument. "The emissions and noise data are overstated" is a different argument, and a weaker one. Conflating the two is how a cost concern gets dressed up as a factual rebuttal it isn't.

What moved the timeline wasn't new science. It was a coalition forming — landscaping workers organizing around occupational exposure, municipal noise complaints reaching a critical mass, battery technology maturing enough that the industry's cost objection lost some of its force. That's the pattern I want to carry into the next section, because I think it's the same pattern now running through AI governance, just on a faster clock.

The Same Pattern Is Running Right Now in AI Governance

If the leaf blower case teaches anything general, it's this: the gap between documented harm and binding rule isn't primarily a knowledge problem. It's a coalition-and-cost problem. Watch how that plays out with AI and the leaf blower story stops looking like a landscaping footnote and starts looking like a preview.

The European Union's AI Act, formally Regulation (EU) 2024/1689, entered into force on August 1, 2024, after years of research documenting algorithmic bias, opaque decision-making, and the risks of automated systems in hiring, credit, and law enforcement. Like California's engine phase-out, the AI Act didn't arrive all at once. Its prohibited-practices provisions became applicable on February 2, 2025. Obligations for general-purpose AI models followed on August 2, 2025. Most obligations for high-risk systems are set to apply from August 2, 2026. That staggered structure is functionally the same choice California made with AB 1346: give the regulated industry a runway instead of an immediate wall, which buys political durability at the cost of speed.

California itself has now applied a version of this pattern to AI directly. Governor Newsom signed the Transparency in Frontier Artificial Intelligence Act, known as SB 53, in September 2025, and it took effect January 1, 2026. Rather than banning anything, it requires developers of the largest frontier AI models to publish safety frameworks and report critical safety incidents. That's a disclosure mandate, a third regulatory tool alongside phase-outs and bans, and it's the tool you reach for when the underlying harm is still hard to define precisely but the information asymmetry between developer and public is the more provable problem.

Underneath both of those binding rules sits something closer to what two-stroke engine science was in the 1990s: evidence-gathering without teeth. The National Institute of Standards and Technology published its AI Risk Management Framework in January 2023, a voluntary guidance document, not a rule. It named the categories of harm, bias, lack of explainability, security vulnerabilities, well before any legislature turned them into obligations. That's the AI equivalent of the engineering literature on two-stroke emissions sitting untouched for decades before AB 1346. The knowledge arrives first. The coalition and the enforceable rule arrive later, and usually only after the cost of inaction becomes harder to wave away than the cost of compliance.

I don't think this parallel is decorative. It's the same institutional reflex showing up in two domains that have nothing else in common. Evidence accumulates quietly for years, largely ignored outside the specialists who study it. Then a jurisdiction moves first, usually not the federal government, and the argument that follows is rarely about whether the underlying harm is real. It's about who absorbs the cost of the fix and how fast they're made to do it.

What This Means If You're Watching How AI Gets Regulated

If you build AI systems, deploy them inside a business, or simply try to reason clearly about where this is heading, the leaf blower case offers a specific, transferable lesson: industry pushback against a rule is not the same signal as a factual rebuttal of the harm the rule addresses, and it's worth training yourself to tell the two apart in real time rather than after the fact.

When you read a company argue that an AI safety disclosure requirement, or a bias-testing mandate, will slow innovation or raise costs, that may be true, and it may still be worth doing anyway. The landscaping industry's objection to AB 1346 was true and real, and California passed the phase-out anyway, because a true cost objection and a false factual rebuttal aren't the same thing, and only one of them is actually an argument against the underlying evidence.

The other transferable lesson is about timing. The jurisdictions that move first on a known harm are rarely the ones with the most resources. They're the ones where the coalition forms first, which in the AI case has visibly been happening at the state level, not the federal level, exactly as it did with leaf blowers. If you're trying to forecast where binding AI rules land next, the leaf blower pattern says: watch the states experimenting with disclosure and phase-in mechanisms now, because that's usually where the federal floor eventually catches up to, years later, once the fact of state-level enforcement makes further delay look like the outlier position rather than the cautious one.

The Open Question I Keep Sitting With

How much evidence does it actually take before a known harm becomes a regulated one? The honest answer, in both of these cases, is that it's rarely about evidence crossing some threshold of certainty. The mechanism was understood decades before the first serious leaf blower statute. The categories of AI harm were named in a NIST framework years before any binding disclosure law existed. What moves is the coalition, not the science.

The case against gas-powered leaf blowers didn't get stronger because someone discovered something new. It got stronger because it kept not going away. I think that's worth remembering the next time an AI harm gets dismissed as overstated. The evidence rarely arrives all at once. It was there all along, waiting for a coalition and a cost curve to catch up to what the facts had already shown.

Frequently Asked Questions

Why did it take roughly a decade for leaf blower evidence to turn into a leaf blower law? Not because the science was uncertain. Two-stroke engine emissions and noise levels were well understood by engineers long before California acted. The delay was a distribution fight over who would absorb the cost of switching equipment and how quickly, and that fight only resolved once battery technology matured enough to weaken the industry's cost objection.

Does AI regulation follow the same phase-out-versus-ban split as leaf blowers? Roughly, plus a third option. The EU AI Act uses a phased compliance timeline similar to California's engine phase-out, giving developers a runway rather than an immediate cutoff. California's SB 53 uses a different tool entirely: a transparency mandate that requires disclosure rather than banning or phasing out any technology. No major AI jurisdiction has yet used an outright use ban comparable to D.C.'s leaf blower ordinance.

Is the lack of a comprehensive federal AI law similar to the lack of a federal leaf blower ban? Yes, structurally. In both cases, the federal government's existing rules (EPA's small-engine emissions standards in one case, a patchwork of sector-specific AI guidance in the other) formalize existing practice rather than push it somewhere new. States moved first on leaf blowers, and states have moved first on binding AI rules as well.

How do you tell when evidence has "accumulated enough" to trigger regulation? There isn't a fixed threshold, and that's the uncomfortable part. What actually triggers a binding rule is a coalition forming around the harm, combined with an alternative becoming cheap enough that the industry's cost objection loses force. Voluntary guidance, like NIST's AI Risk Management Framework or early manufacturer noise disclosures, usually arrives years before the coalition does.

What should someone building or deploying AI take from the leaf blower case? Treat accumulating evidence about AI harms as a forecast, not a settled debate to win or lose. And separate a company's legitimate cost objection to a proposed rule from a claim that the underlying harm isn't real. Those are two different arguments, and only the second one actually challenges the evidence.

Last updated: 2026-08-29