AI bots flooded UK vape ban consultation with 90,835 fake responses

Bots submitted 90,835 of the 118,756 responses — 76% — to the UK government's 2023 consultation on banning disposable vapes, in what researchers call the first documented case of large-scale automated interference in a formal public health policy consultation.

The attempt failed. The Department of Health and Social Care (DHSC) detected the machine-generated submissions, excluded them from its decision-making, and the ban on selling or supplying disposable vapes became law anyway. But the case, published Tuesday in the journal Tobacco Control, exposes a structural weakness in how governments gather public input — and how cheaply that process can now be attacked.

The research was led by Dr Nason Maani of the University of Edinburgh and Prof Anna Gilmore of the University of Bath, two health policy experts who documented the bot campaign and the government's response.

What actually happened?

The flood targeted the "smoke free generation" consultation run by Rishi Sunak's Conservative government in 2023, part of the groundwork for the tobacco and vapes bill. All but two of the 90,835 automated responses answered the same question, and every one of them opposed the plan to ban disposable vapes.

The submissions were not sophisticated. According to the DHSC, they were "often duplicated many times and sent at high frequency from a small number of IP addresses." That sloppiness is precisely what made them catchable. Bot-detection methods flagged the campaign, and the department removed the fake responses before they could shape the outcome.

The researchers wrote that the bombardment "appears to be the first documented instance of large-scale automated interference in a formal public health policy consultation." The ban survived this time because the fakes were clumsy and the controls held.

Who was behind it?

The research published in Tobacco Control does not name a culprit, and the source material contains no attribution to a specific actor. What it does establish is the pattern: tens of thousands of near-identical responses, one target question, one policy objective — stopping the disposable vapes ban.

Caroline Cerny, deputy chief executive of Action on Smoking and Health, pointed to the obvious suspect class. "Given the tobacco industry's history of seeking to undermine public health regulation, we need to be alert to the risk that technological advances can be used to further commercial interests around health policy and put safeguards in place," she said.

She added: "Decisions that affect public health must be based on sound evidence and expertise, not manipulated response numbers."

Why the detection worked this time

The DHSC has defended its process. A spokesperson said: "We always have robust controls in place to prevent the manipulation of our public consultations, and those controls worked effectively in this case."

The spokesperson continued: "As a result, the fake responses were removed, and the consultation helped inform the landmark action we are taking to create a smoke-free generation. We will remain alert to interference of this kind in future consultation processes, and we will not be deterred from doing what is necessary to promote healthier living."

The mechanics matter here. Duplicate text and a small number of IP addresses are easy signals to catch. Generative AI erodes both. A modern campaign can produce unique, plausible-sounding responses at scale and distribute them across many addresses, which removes the very fingerprints that exposed this attempt. Detection that worked in 2023 is not guaranteed to work against the next campaign.

That is Maani's core warning. While this attempt was identified, he said, "the threat is that future campaigns become harder to catch, and what we can't catch, we don't know about. Such campaigns are a threat because they raise the possibility of a handful of powerful interests looking like a mass movement."

That sentence is the stakes of the whole story: a consultation exists to measure what the public thinks. If a small group can fabricate the appearance of a mass movement, the instrument stops measuring anything real.

What did the bill do?

The policy the bots tried to kill is now law. Keir Starmer's Labour government inherited the tobacco and vapes bill in 2024 and passed it, including the ban on businesses selling or supplying disposable vapes in the UK.

So the episode ends with the manipulators losing on the substance. The consultation served its purpose once the fake responses were stripped out, and the legislation proceeded. The outcome is a clean win for the policy process in this single case — and a warning light for every consultation that follows.

Why there is no easy fix

Azzurra Moores, associate director for information ecosystems at the thinktank Demos, described the attempt as "pretty alarming" and laid out the dilemma departments now face.

"Public consultations are one of the main ways government opens up its decision-making, so any interference with them is a serious concern," she said. "If a consultation can be swamped like this, it stops telling policymakers what the public actually thinks and undermines trust in the process."

Her diagnosis of the defensive problem is blunt: "This is a problem without easy answers. Every barrier that keeps bots out also makes it harder for real people to take part, and AI makes fake responses far harder to spot. Interference can adapt faster than departments can update their checks. Better screening and more transparency will help, but there's no single fix."

That asymmetry is the structural issue. Raising the barrier to entry filters out bots but also filters out ordinary citizens. Lowering it invites exactly the kind of flood the DHSC caught. And the attackers can iterate faster than a government department can revise its verification procedures.

Cerny's framing matches Moores's on the trust dimension. "Evidence that AI bots have been used to swamp a Department of Health consultation is deeply concerning," she said. Her organization's position is that the tobacco industry's record of fighting regulation makes technological manipulation of health policy a live risk, not a hypothetical one.

What this means for consultations everywhere

The UK case is documented, caught and survived — which makes it unusually informative. Most attempted manipulations that succeed never surface, because a successful campaign looks like genuine public opinion. Researchers only get to study the failures.

For policymakers, the takeaway from Tobacco Control is procedural: consultation systems need bot detection as standard infrastructure, not as an afterthought. For the public, it is a caveat: response counts in any contested consultation can now be manufactured, and a headline number of submissions is no longer evidence of public sentiment on its own.

Maani's warning sets the agenda for what comes next. The 2023 campaign failed because it was crude and the DHSC's controls held; the next one may not be crude, and the controls that caught 90,835 duplicates will not necessarily catch a campaign designed to look like 90,835 different people.