The UK's Generational Tobacco Ban Is a Blueprint for How We Should Regulate AI in 2026

The UK's generational tobacco ban — which prevents anyone born after 2009 from ever legally buying cigarettes — might fail on its own terms. But the thinking behind it is arguably the most important regulatory philosophy of our era, and it's one that AI policymakers are almost completely ignoring.

There's a journalist at MIT Technology Review who supports the tobacco ban despite its uncertain odds of success. She's a parent. Her kids are repulsed by smoking in a way her generation simply wasn't. And that cultural shift, she argues, is worth legislating around even if enforcement is imperfect. I'm not here to rehash her argument. I'm here to ask the obvious follow-up question that the AI industry desperately needs to confront: if we're willing to make generational bets on tobacco, why aren't we making them on AI?

The Generational Harm Framework — And Why AI Needs It Urgently

The tobacco ban's underlying logic is elegant and underappreciated. It doesn't try to eliminate smoking overnight. It doesn't criminalise existing smokers. It draws a line in the sand and says: from this cohort forward, we are not going to let commercial interests normalise something we know causes harm. The policy accepts short-term messiness — black markets, enforcement gaps, civil liberties debates — in exchange for a long-term cultural and health trajectory that bends away from damage.

Now consider AI. The seven-year-old learning about AI at school today will be 21 in 2040. The five-year-old getting internet-based homework will be entering the workforce around the same time. These children are not hypothetical future users. They are being shaped right now by systems whose long-term cognitive, social, and economic effects we genuinely do not understand. And unlike tobacco — where we had decades of epidemiological data before serious regulation — we are regulating AI (or failing to) in real time, while the experiment runs on actual children.

The EU AI Act, the US Executive Orders, the UK's own rather toothless pro-innovation AI framework — none of them deploy the generational harm logic that the tobacco ban uses. They regulate use cases, risk tiers, and liability frameworks. That's necessary but insufficient. What they don't do is ask: what kind of AI relationship do we want the next generation to have, and are we actively designing toward that?

What "Generational Thinking" Would Actually Look Like in AI Policy

Let's get concrete. A generational approach to AI regulation wouldn't ban large language models or algorithmic recommendation systems. It would instead identify specific interaction patterns — compulsive engagement loops, AI companionship substituting for human relationships, algorithmic content shaping political identity during adolescence — and draw lines around those for younger cohorts.

Think about what that means in practice. Age-verified AI companion apps that are legally prohibited from forming emotionally dependent relationships with users under 18. Recommendation algorithms that, for users born after 2015, are required to operate on fundamentally different optimisation targets — diversity of viewpoint over engagement maximisation, for instance. AI tutoring systems that must, by law, periodically hand cognitive load back to the student rather than solving problems on their behalf.

None of these are technically impossible. Several are already being explored voluntarily by more conscientious developers. The question is whether we have the regulatory will to make them mandatory and generationally binding — to say, in effect, "this cohort deserves a different relationship with AI than the one we accidentally built for ourselves."

The Enforcement Problem Is Real — And It's Not the Point

Critics of the tobacco ban love pointing out that it'll be unenforceable. Kids will get cigarettes. Black markets exist. And they're probably right, at least partially. The Technology Review writer's response — that she supports it anyway — is not naïve. It's a sophisticated position about what law is actually for.

Law doesn't only function through enforcement. It functions through norm-setting. The UK's tobacco ban tells a generation of children something about how their society values their lungs. It changes what parents say, what schools teach, what advertisers can do. The enforcement gap is a bug; the cultural signal is the feature.

AI regulation suffers from exactly the inverse problem. We have plenty of enforcement mechanisms being built — the EU's fines, the FTC's growing teeth, the emerging liability frameworks for AI-generated harm. What we almost entirely lack is the cultural signal. There is no regulatory statement that tells today's children something coherent about how their society values their minds, their attention, their cognitive autonomy, or their right to develop beliefs that weren't shaped by a revenue-optimised model.

That absence is not neutral. It is itself a choice, and it is one that will compound for decades.

What This Means for Developers and Businesses Right Now

If you're building AI products in 2026 — and especially if any part of your user base skews young — the generational harm framework is coming for your industry whether you prepare for it or not. Here's what forward-thinking teams should be doing today:

Document your design choices around young users explicitly. Not just for compliance, but because regulators who adopt generational thinking will ask for exactly this paper trail. What did you know, when did you know it, and what did you change?

Invest in longitudinal research partnerships. The tobacco industry's catastrophic credibility failure came partly from suppressing long-term data. AI companies that proactively fund independent, long-term studies into cognitive and social effects on young users will be in an entirely different regulatory and reputational position in 2035.

Build age-differentiated product architectures now, not later. Retrofitting is always more expensive and more disruptive than designing for it upfront. The companies that treat "how does this product behave differently for a 14-year-old versus a 40-year-old" as a first-class design question will have a significant advantage when the regulation inevitably arrives.

The UK's tobacco ban might not work perfectly. But the instinct behind it — that some harms are serious enough to justify drawing a line for the next generation, even at cost, even with uncertainty — is exactly the instinct that AI regulation is currently missing. The children learning about AI at school today deserve more than an industry that's still figuring out its own terms of service. They deserve a society that made a bet on their behalf.