A public consultation meant to capture human views on disposable vapes was instead flooded with automated submissions, according to an analysis published in a leading medical journal. Researchers say the episode represents the first documented case of large-scale bot interference in a formal public health consultation, and it offers a preview of how AI can be used not merely to spread false claims but to distort the procedural machinery through which health policy is made.

The consultation, run by the Department of Health and Social Care, asked the public for opinions on disposable vapes and related health questions in 2023. It received 118,756 responses. When researchers examined the submissions, they found a pattern that looked less like a groundswell of public opinion and more like a coordinated campaign. Roughly 76% of the responses were eventually flagged as fraudulent. Of those, 90,833 out of 90,835 answered only one question — the one asking whether disposable vape sales should be restricted. The answers overwhelmingly opposed restrictions.

Key facts at a glance

• The consultation drew 118,756 public responses in a multiple-choice format.
• About 76% of responses were flagged as fraudulent.
• 90,833 of 90,835 suspicious responses answered only the question about restricting disposable vape sales.
• Bot-generated responses overwhelmingly opposed restrictions on disposable vapes.
• After fraudulent responses were removed, 79.6% of remaining respondents favored restrictions.
• Researchers described it as the first documented instance of large-scale automated interference in a formal public health consultation.
• The pattern resembles earlier mass coordinated lobbying by tobacco interests, but with far lower costs and greater scale.

The striking detail is not simply the volume of suspicious responses. It is their narrow focus. A genuine respondent to a broad health consultation might answer several questions, skip others, or offer a mix of views. The automated submissions, by contrast, concentrated almost entirely on a single policy question. That directional consistency — overwhelmingly opposing restrictions — suggests a targeted effort rather than an organic outpouring. Researchers said the scale and consistency pointed to a coordinated campaign, consistent with prior evidence that tobacco industry allies have used mass coordinated submissions to shape public consultations.

Why public consultations matter

Public consultations are a standard tool of modern governance. They allow agencies to hear from affected communities, identify unintended consequences, and build legitimacy for decisions that may be controversial. They are not referendums, and they are rarely the only source of evidence. But they can influence how regulators frame a problem, how quickly they act, and how much political cover they have. That makes them a tempting target for anyone who wants to delay, weaken, or discredit a policy.

Health policy is especially vulnerable because the stakes are high and the controversies are durable. Tobacco control, alcohol regulation, sugar taxes, vaccine requirements, drug pricing, and environmental health rules all attract well-funded opposition. In the past, influence campaigns required lobbyists, PR firms, and networks of sympathetic organizations. AI changes the economics. A single operator can generate thousands of tailored comments, each written in fluent prose, each appearing to come from a different citizen. The cost is low, the output is fast, and the text can be varied enough to evade simple duplicate detection.

The disposable vape dispute fits that pattern. Governments have been under pressure to restrict single-use vapes because of concerns about youth nicotine addiction, environmental waste, and the ease with which disposable products can be purchased. At the same time, some adults use vapes as a smoking cessation tool, and industry groups argue that broad bans could push former smokers back to cigarettes or create illicit markets. Those are legitimate policy arguments. The problem arises when a campaign tries to simulate public consensus instead of making its case openly.

How the manipulation was detected

The suspicious pattern emerged from the structure of the responses. The consultation used multiple-choice questions, which makes it easier to analyze at scale. Researchers looked at how many questions each respondent answered, how consistent the answers were, and whether the submissions clustered around a single issue. The concentration was extreme: nearly all of the flagged responses answered only the disposable vape question. That is not how a typical public consultation looks. It is how a script looks.

The researchers could not identify who was behind the campaign. That is one of the most troubling aspects of the case. The submissions were designed to appear human. They may have been generated by language models, automated form-filling tools, or a combination of both. Without metadata, device fingerprints, or admissions from the operators, attribution is difficult. The episode therefore does not prove that a specific company or industry group was responsible. It does show that the technique works, that it can be executed at scale, and that current consultation systems are not reliably equipped to stop it.

The first-documented label should also be treated cautiously. It means researchers found and analyzed a clear case, not that earlier attempts never happened. Many consultations are analyzed quickly, with limited forensic resources. Automated manipulation can blend into ordinary internet noise, especially when agencies are more focused on summarizing views than on verifying identities. The true prevalence is unknown, and that uncertainty is part of the risk.

The bigger AI threat to policymaking

Much of the public debate about AI and health has focused on dramatic harms: chatbots giving dangerous medical advice, models being misused to design pathogens, or automated systems making biased decisions about care. Those risks are real. But the vape consultation points to a quieter and more immediate danger. AI can be used to manipulate the inputs of democratic decision-making — the comments, consultations, petitions, and public signals that officials use to judge what the public wants.

This is sometimes called participatory manipulation. It does not require hacking a government database or breaking into a secure system. It exploits open processes that are designed to be accessible. The more open the process, the more vulnerable it may be. Agencies want to hear from as many people as possible, so they keep barriers low. Bad actors can take advantage of that openness by submitting fake responses in bulk. If agencies respond by raising barriers, they may exclude the very people they want to hear from. If they do nothing, they risk letting fabricated sentiment shape policy.

The vape case also shows how AI can blur the line between lobbying and impersonation. Traditional lobbying is transparent in principle: organizations register, disclose their interests, and argue in public. Automated submissions pretend to be citizen voices. That is a different kind of intervention. It does not just add a perspective to the debate; it counterfeits the public. When that happens, the consultation no longer measures public opinion. It measures who can generate the most convincing synthetic noise.

What regulators can do

There is no single fix. Consultation systems need layered defenses. Identity verification can help, but it must be designed to protect privacy and avoid excluding vulnerable groups. Rate limits can slow bulk submissions, but they can be bypassed. CAPTCHAs can filter simple bots, but they are increasingly ineffective against AI-assisted tools. Statistical anomaly detection can flag suspicious patterns, but it needs skilled analysts and clear thresholds. Text provenance tools and watermarking may help in some cases, but they are not yet reliable enough to serve as a sole gatekeeper.

Process design matters as much as technology. Agencies can separate open public comment from representative sampling. They can use randomly selected citizen panels to complement open consultations. They can publish submission counts, methodology, and anomaly reports so that outside researchers can audit the process. They can require disclosure when submissions are generated with AI assistance, while still allowing legitimate advocacy. They can also move away from treating raw comment counts as a measure of public support. A million identical submissions should not carry the same weight as a well-designed survey or a detailed submission from a community group.

Detection will always be a moving target. As language models improve, synthetic comments will become harder to distinguish from human ones. The better response is to strengthen the integrity of the process rather than rely on a single detection tool. That means better data hygiene, clearer rules about coordinated campaigns, and more transparency about how officials weigh public input. It also means treating consultation fraud as a governance problem, not just a cybersecurity problem.

What the vape case signals

The disposable vape consultation is a small case in one policy area, but its implications are broad. It shows that AI can be deployed to sabotage public health policies without ever making a public argument. The goal is not necessarily to convince officials with better evidence. It is to create the appearance of opposition, to overwhelm the signal, and to make the process look illegitimate. Even if the effort fails, it can sow doubt about whether any consultation reflects real public sentiment.

After the fraudulent responses were removed, 79.6% of the remaining respondents favored restrictions on disposable vape sales. That reversal is important. It suggests that the automated campaign did not reflect the underlying balance of opinion among genuine participants. But the fact that the campaign was detected does not mean the next one will be. The researchers could not trace the origin of the submissions, and the tools used to generate them are becoming cheaper and more accessible. Governments, health agencies, and civic institutions are already warning about a coming wave of AI-enabled attacks on public processes. The vape consultation offers an early look at what that wave may look like: not a dramatic hack, but a flood of fluent, plausible, fake citizens.

The episode also complicates the common assumption that more public participation is always better. Participation is valuable when it is authentic. When it can be manufactured at scale, agencies need new ways to distinguish between a public voice and a public relations product. The policy question is no longer just whether disposable vapes should be restricted. It is whether the public consultation itself can survive as a trustworthy instrument of health governance.

Source:Digital Trends News