Key facts

• Researchers have documented what they describe as the first large-scale automated interference in a formal public health consultation.
• The case involved a 2023 consultation on disposable vapes and other tobacco-related policies.
• Of 118,756 responses, about 76% were eventually flagged as fraudulent.
• An estimated 90,833 of 90,835 responses answered only one question — whether to restrict disposable vape sales.
• Responses generated by bots overwhelmingly opposed restrictions on disposable vape sales.
• After bot-generated responses were removed, 79.6% of remaining respondents favored restrictions.
• The pattern suggested a coordinated campaign, echoing earlier tobacco industry tactics.
• Researchers could not identify who was behind the submissions.

Public health policymaking depends on the assumption that the people who submit comments to government consultations are people. That assumption is now under strain. A newly documented case shows how automated accounts and AI-generated text can be used to flood a formal health consultation, mimic public opinion, and push back against rules designed to protect citizens. The target in this case was disposable vapes, a product at the center of a heated debate over youth nicotine use, environmental waste, and smoking cessation.

The case centers on a 2023 consultation in which the U.K. Department of Health and Social Care asked for views on disposable vapes and related tobacco policies. Members of the public were invited to answer multiple-choice questions, a standard way for governments to gauge support for proposed rules. The consultation drew 118,756 responses. That volume alone would have been notable. What followed was more troubling: researchers later concluded that about 76% of the submissions were fraudulent, with automated activity responsible for a large share.

The most striking detail involved a single question. Of 90,835 responses that addressed whether disposable vape sales should be restricted, 90,833 answered only that question. They did not engage with the rest of the consultation. The responses were also overwhelmingly opposed to restrictions. That combination — an enormous wave of single-question submissions, all leaning in the same direction — raised suspicion. When researchers analyzed the submissions, they found patterns consistent with bot-generated text rather than individual human participation.

After the fraudulent responses were removed, the picture changed dramatically. Among the remaining responses, 79.6% favored restrictions on the sale of disposable vapes. In other words, the raw consultation data suggested strong opposition to regulation, while the cleaned data showed strong support. The gap was not a minor discrepancy. It was a reversal large enough to change how a policymaker might read public sentiment.

Researchers described the episode as the first documented instance of large-scale automated or bot interference in a formal public health consultation. That distinction matters. Public consultations are not opinion polls in the traditional sense, but they are used as evidence of public concern, stakeholder preference, and democratic legitimacy. If a coordinated campaign can manufacture thousands of fake responses, it can distort that evidence and create a false impression of grassroots opposition.

The pattern also has historical echoes. The scale and directional consistency of the bot responses suggested a coordinated campaign, researchers said, consistent with prior evidence of tobacco industry efforts to manipulate public consultations through mass coordinated submissions. For decades, tobacco companies and their allies have been accused of using front groups, letter-writing campaigns, paid consultants, and other tactics to slow regulation. Automated text generation adds a new layer of speed and scale to that playbook.

Disposable vapes are a high-stakes target for this kind of interference. Health officials have raised alarms about youth vaping, nicotine addiction, and the environmental impact of single-use devices. At the same time, some adults use vapes as a smoking cessation tool, and regulators must weigh those benefits against risks. The policy debate is emotional and politically sensitive. That makes public consultations valuable to both sides — and it makes them vulnerable to manipulation. A campaign that can tilt the visible response count can shape media coverage, political messaging, and the perceived legitimacy of new rules.

The mechanics of AI-generated interference are not especially exotic. Large language models can produce fluent, grammatically correct text at scale. Automation scripts can submit forms repeatedly. When a consultation allows open-ended comments or multiple-choice answers, attackers can generate responses that look plausible enough to pass casual review. Detection often depends on statistical anomalies: identical phrasing, repeated answer patterns, unusual timing, or responses that address only one question while ignoring the rest. In this case, the concentration of single-question answers and the overwhelming directional bias helped expose the campaign.

The implications extend beyond vaping. Governments increasingly use digital consultations to gather input on food policy, drug regulation, vaccination, environmental health, and emergency preparedness. Each of those areas could become a target. A coordinated bot campaign could flood a consultation on vaccine policy with comments opposing mandates, or a consultation on air quality with comments opposing emission limits. Even if the fake responses are eventually removed, they can create doubt, delay action, and force agencies to spend time and money on verification. The damage is not only to the final count but to public trust in the process.

The response to this threat will require more than better spam filters. Consultation platforms may need identity verification, rate limiting, anomaly detection, and independent audits. Agencies may need to publish not only raw response totals but also the results after fraud screening. Researchers and civil society groups may need access to anonymized data so they can check for manipulation. And policymakers may need to treat public consultations as one input among many rather than a direct plebiscite. None of these measures is simple. Privacy concerns, accessibility, and the risk of excluding legitimate participants all complicate the design.

There is also a broader information problem. AI-generated text is cheap, fast, and increasingly convincing. The same tools that can help a patient understand a diagnosis or help a student draft an essay can also be used to manufacture the appearance of public support. In the vaping consultation, the fake responses were not sophisticated enough to fool researchers after detailed analysis. But they were apparently enough to represent a large share of the raw submissions. That is a warning: detection may work in retrospect, but the political narrative can be shaped before the cleanup happens.

The researchers could not pinpoint the origin of the malicious responses. That unanswered question leaves the episode open to speculation. Was it a commercial actor with a financial interest in disposable vapes? A political campaign? A foreign influence operation? A prank that got out of hand? The analysis did not identify a source. What it did identify was a method and a vulnerability. The method is automated text generation combined with mass submission. The vulnerability is a consultation system designed for human participation but not built to withstand industrial-scale deception.

Public health agencies have already warned that AI could be misused in ways that range from disinformation to bioweapon research. The vaping consultation offers a more mundane example: not a catastrophic attack, but a targeted effort to distort a policy process. That may be how AI interference often appears in practice — not as a single dramatic event, but as a series of smaller manipulations that erode confidence in institutions. The first documented case is now on the record. The question is how many others have gone undetected.

For now, the numbers tell a clear story. A consultation that appeared to show strong opposition to disposable vape restrictions actually showed strong support once automated submissions were removed. The difference was not a matter of interpretation. It was a matter of fraudulent data. The episode has been described as the first documented case of large-scale bot interference in a formal public health consultation, and it has moved the issue from speculation to evidence. The researchers have not identified the origin of the fraudulent submissions, leaving a central question unanswered as governments prepare more consultations on health policy.

Source:Digital Trends News