Edinburgh study finds 90,000 AI bots flooded UK smoking ban consultation

A University of Edinburgh study published in the journal Tobacco Control on Tuesday found that AI bots submitted 90,000 of the 118,756 responses to a government consultation on a lifetime smoking ban. The automated entries represented 76% of the total submissions received in October 2023.

The consultation preceded the passage of the Tobacco and Vapes Bill, which became law in April. The legislation bans the sale or supply of tobacco to anyone born in 2009 or later to create a smoke-free generation. It also restricts vapes, prohibiting online and in-store sales to those 17 and younger. The law will ban “confectionery” flavors, such as gummy bears and cotton candy, which health experts say target children.

Lead author Dr. Nason Maani, a senior lecturer in inequality at the University of Edinburgh, stated that the bot submissions were primarily opposed to the vape restrictions. His team identified the automated entries through Department of Health and Social Care data. The submissions arrived in a high-frequency wave from a small number of IP addresses. They contained almost identical multiple-choice answers that did not follow the survey flow of genuine respondents.

Dr. Maani noted that the bots did not rely on common AI hallmarks like em dashes or emojis. Instead, the mass duplication and speed of the submissions indicated a coordinated campaign. If the Department of Health and Social Care had not filtered out the fake responses, the raw data would have shown 81% opposition to disposable vape restrictions. Among verified respondents, roughly four in five supported the measures.

The study suggests the tactic mirrors historical strategies used by tobacco lobbyists to sway public opinion. Dr. Maani said the pattern is consistent with previous efforts to swamp consultations, though the data does not identify the specific actors involved. He warned that consultations inform policy, making them vulnerable to manipulation when automated responses are not detected.

The Department of Health and Social Care stated that its controls worked effectively in this instance. The department removed the fake responses and used the verified data to inform its regulatory actions. Officials said they would remain alert to similar interference in future processes.

Dr. Maani said the discovery serves as a reminder of the need for deliberative approaches and live participation in public consultations. These methods are harder to automate than digital submission portals. The study appears in the journal Tobacco Control.