Key facts

• A public health consultation on disposable vapes received 118,756 responses.
• About 76% of the submissions were later flagged as fraudulent.
• Of 90,835 responses that answered only one question, 90,833 addressed restrictions on disposable vape sales.
• The suspicious responses overwhelmingly opposed restrictions.
• After bot-like submissions were removed, 79.6% of remaining respondents supported restrictions.
• Researchers describe the case as the first documented large-scale automated interference in a formal public health consultation.
• The pattern resembles coordinated tobacco-industry lobbying, and the origin of the submissions remains unknown.

AI-generated pressure on a public health debate

In the broad debate over artificial intelligence, health and medical risks often dominate the conversation. Much of that attention has focused on dramatic scenarios, including warnings from frontier AI developers that powerful models could be misused to create bioweapons or spread dangerous medical misinformation. Security experts have already identified incidents in which AI systems without strong guardrails were used in harmful ways. The disposable vape consultation case points to a less theatrical but equally troubling risk: automated systems can imitate public opinion at scale and attempt to shape the rules that govern public health.

The episode centers on a 2023 public consultation that invited people to comment on disposable vapes and other health topics. When researchers analyzed the submissions, they found evidence that a large share of the responses had been generated by bots rather than written by individual citizens. Suspicion grew because the responses were so heavily weighted in one direction. Instead of reflecting a normal mix of views, the submissions overwhelmingly opposed restrictions on disposable vape sales. That pattern was unusual enough to prompt a closer examination of who, or what, was responding.

How the consultation was flooded

The consultation asked for public input through multiple-choice questions, and it attracted 118,756 responses. That is a substantial number for a health policy consultation, but volume alone does not prove manipulation. The more revealing signal was the structure of the responses. Researchers found that 90,833 out of 90,835 responses answered just one question. That single question sought opinions on restricting disposable vapes. In other words, thousands of submissions appeared to be designed for one narrow purpose: to create the impression of broad public opposition to a vape crackdown.

Eventually, 76% of the submissions were flagged as fraudulent. The researchers did not conclude that every flagged response was produced by an advanced language model, but they described the activity as automated or bot-driven interference. The submissions were dressed up as the views of real people, which is what makes the case so significant. A public consultation is supposed to help officials weigh competing interests, assess public sentiment, and identify policy consequences. When automated accounts flood that process, they do not simply add noise. They can distort the evidence base that policymakers use to justify or reject health regulations.

The distortion becomes clearer when the suspicious responses are removed. Once the bot-like submissions were discarded, the remaining responses shifted sharply. A 79.6% majority of respondents supported implementing restrictions on the sale of disposable vapes. That result is nearly the opposite of the impression created by the raw submission data. The raw numbers suggested strong opposition to restrictions. The cleaned data suggested strong support. For public health officials, that gap is not a technical curiosity. It is a direct challenge to the integrity of the consultation process.

Why disposable vapes became a battleground

Disposable vapes have become a major public health issue because they sit at the intersection of several concerns. Health officials worry about youth nicotine use, addiction, and the appeal of flavored products. Environmental agencies worry about the waste created by single-use batteries and plastic components. Parents, schools, and medical groups have pressed for tighter rules, while retailers and some consumers argue that restrictions could reduce options for adults trying to switch away from cigarettes. The debate is intense because the stakes are not abstract. Regulations can affect what products are sold, how they are marketed, and how quickly young people can access them.

That intensity makes disposable vapes a natural target for coordinated influence campaigns. Public consultations are designed to gather evidence from a wide range of people, including health professionals, industry representatives, consumers, and community groups. But they depend on a basic assumption: that each submission represents a genuine voice. Automated responses break that assumption. They can be produced cheaply, repeated across thousands of accounts, and tailored to sound like ordinary citizens. The result is a form of synthetic public opinion that can be used to argue that restrictions are unpopular, even when the underlying public is more supportive.

The lobbying playbook, automated

The researchers noted that the manipulation effort resembled earlier lobbying tactics used by the tobacco industry and other corporate actors. In the past, coordinated campaigns have encouraged supporters to submit mass responses to consultations, sometimes using template letters or prewritten arguments. Those campaigns were already a concern because they could create the appearance of grassroots support. AI-generated bots take the same playbook and make it faster, cheaper, and more difficult to detect. Instead of relying on human participants to copy a template, automated systems can generate unique-sounding submissions at scale.

That shift matters for public health policy. Tobacco industry lobbying has a long history of challenging regulations, questioning evidence, and emphasizing economic or personal-choice arguments. The disposable vape consultation showed a similar directional pattern: the suspicious responses were heavily opposed to restrictions. The researchers described the scale and consistency of the bot responses as suggestive of a coordinated campaign. They did not identify who was behind the submissions, but they pointed out that the pattern was consistent with prior evidence of organized efforts to manipulate public consultations through mass coordinated submissions.

A first-of-its-kind warning

The case is described as the first documented instance of large-scale automated or bot interference in a formal public health consultation. That distinction is important because it moves the discussion from speculation to evidence. For years, experts have warned that AI could be used to manipulate democratic processes, overwhelm public agencies, or manufacture false consensus. Much of that debate has focused on elections, social media, and financial fraud. The vape consultation shows that health policy is also exposed. Public health agencies make decisions that affect millions of people, and those decisions often rely on consultations that are vulnerable to automated manipulation.

The researchers could not pinpoint the origin of the malicious responses. That uncertainty is itself a warning. Automated campaigns can be launched from many locations, routed through different networks, and designed to obscure their source. Even when the pattern is detected, attribution can be difficult. The absence of a clear culprit does not make the interference less harmful. It makes it harder to deter. If agencies cannot identify who is manipulating a consultation, they cannot easily impose consequences or prevent the same actors from trying again.

Why public consultations are vulnerable

Public consultations are built on trust. They assume that respondents are who they say they are, that they are responding independently, and that their answers reflect genuine preferences. These assumptions are not always stated, but they are essential to the legitimacy of the process. When an agency receives a large number of responses, it may treat the volume as evidence of public concern. When the responses are generated by bots, that evidence becomes unreliable. The agency may end up weighing synthetic opinion as if it were real.

Detection is not simple. Automated text can be fluent, varied, and plausible. A poorly designed bot campaign might reveal itself through duplicate phrasing or identical answers, but a more sophisticated operation can avoid those obvious signs. Researchers and security teams must look for patterns across the entire dataset. They may examine response timing, completion rates, metadata, account behavior, and consistency across questions. In the vape consultation, the concentration of responses on a single question was a major red flag. That kind of structural anomaly can be more revealing than the wording of any individual submission.

The challenge is also an arms race. As detection tools improve, bad actors can adapt. They can vary language, mimic local concerns, and spread submissions over time. Public agencies may not have the resources or technical capacity to police every consultation at the level required. That leaves health policy exposed to a form of manipulation that is cheap to execute and difficult to prove in individual cases. The vape consultation is a warning that the problem is not theoretical. It has already happened in a formal process with real policy consequences.

Health policy and democratic trust

If bot submissions can sway or delay health regulations, the effects go beyond vaping. Public health policy depends on public trust. People need to believe that decisions are based on evidence, that consultations are fair, and that officials are listening to genuine concerns. When automated submissions distort that process, they can undermine confidence in the outcome. Supporters of restrictions may feel that their voices were drowned out. Officials may face pressure based on manufactured opposition. The resulting policy may be weaker, slower, or more confused than it would have been otherwise.

Agencies can take steps to reduce the risk. They can require stronger identity verification, use rate limiting, analyze submission patterns, and sample responses for authenticity. They can separate open comment periods from structured survey questions and use random sampling to estimate public opinion more reliably. They can also publish transparency reports explaining how submissions were screened and how many were excluded. None of these measures is perfect, but together they can make automated manipulation harder and more costly. The alternative is to accept that any public consultation can be flooded by synthetic voices.

The wider AI risk landscape

The vape consultation fits into a broader set of concerns about AI misuse. Frontier AI developers have warned that powerful models could be used to create bioweapons, generate dangerous medical advice, or spread misinformation at scale. Security experts have already spotted incidents involving models without solid guardrails. Banks, governments, and corporate entities around the world have warned about an onslaught of AI-powered attacks in the near future. The common thread is scale. AI can lower the cost of producing convincing content and make it easier to target institutions that rely on public input.

Not every AI risk involves a dramatic attack. Some of the most damaging uses may be bureaucratic, quiet, and repetitive. Flooding a consultation with fake responses is less visible than a cyberattack, but it can have a similar effect on institutional decision-making. It exploits the gap between the volume of input an agency can collect and its capacity to verify that input. As AI tools become more accessible, that gap is likely to widen. Agencies that rely on public consultations will need to treat submission integrity as a core security issue, not an administrative afterthought.

What remains unresolved

The researchers could not identify the origin of the bot-generated responses, and no one has established a definitive link to a particular industry or campaign. That leaves important questions unanswered. Were the submissions produced by a small group with technical skills, a hired influence operation, or a broader coordinated network? Did the operators understand the policy process, or were they simply testing a tool? Will the same tactics appear in other health consultations, including those on tobacco, alcohol, food labeling, or environmental health? The case provides evidence of a technique but not a complete account of the actor behind it.

What is clear is that the vape consultation changed the factual record. The raw submissions created an impression of overwhelming opposition to restrictions. The cleaned data showed a strong majority in favor. That difference is the core of the story. It shows how automated systems can manufacture a version of public opinion that looks real enough to influence policymakers. It also shows why agencies need to adapt before the next consultation is flooded. The question is no longer whether automated systems can imitate public opinion at scale. The consultation showed that they can. The harder question is how institutions will verify who is speaking before the next health policy is written.

Source: Digital Trends News