AI-Generated Public Submissions in EU Tobacco Regulation Consultation
In May 2026, Philip Morris International (PMI) reportedly deployed an AI-powered tool named Your Voice Your Choice to generate public submissions opposing stricter EU tobacco regulations. The tool was used during the European Commission’s call for evidence on proposed revisions to the Tobacco Products Directive and the Tobacco Advertising Directive. Investigations by European journalists, including NOS and the European Investigative Collaborations (EIC), revealed that hundreds—potentially thousands—of submissions were AI-generated, raising concerns about the integrity of public consultations and the erosion of democratic processes.
The European Commission’s consultation, which ran from late 2025 to early 2026, received nearly 90,000 responses. Journalistic analyses indicated that a significant portion of these submissions, particularly in the Netherlands, were traced back to the AI tool. In some cases, nearly three-quarters of Dutch submissions were fully AI-written, presented as individual citizen responses rather than corporate or industry-backed input. This scale of AI-mediated participation suggests an attempt to manipulate public opinion and regulatory outcomes through automated, mass-generated content.
The tool allegedly guided users toward arguments opposing stricter tobacco rules, even when their initial selections did not align with such positions. This behavior undermines the transparency and fairness of public consultations, where stakeholders are expected to engage authentically rather than through algorithmic amplification of pre-determined narratives.
The AI system in question appears to function as an AI-assisted public consultation response generator, likely built on large language models (LLMs). Such tools are designed to automate the creation of personalized submissions, but their deployment in regulatory processes raises ethical and legal questions. The European Commission has not yet formally addressed the incident, though the allegations have sparked debates about the need for stricter oversight of AI in public policy-making.
Mechanism and Scale of AI-Mediated Influence
The Your Voice Your Choice tool operated by allowing users to input basic preferences or concerns about tobacco regulation, which the AI then expanded into fully formed submissions. However, investigations found that the tool did not merely assist in drafting responses—it actively steered content toward industry-friendly positions. For instance, users who selected options supporting stricter advertising bans or youth protection measures may have received submissions that included counterarguments, such as claims about the economic impact of regulations or the role of vaping in harm reduction.
The scale of AI-generated submissions varied by country, with the Netherlands particularly affected. The findings suggested that the tool was used systematically to flood the consultation with pro-industry narratives, creating the illusion of broad public opposition to regulatory changes. This tactic is reminiscent of astroturfing—where artificial grassroots support is manufactured to influence policy—exacerbated by the use of AI to automate the process at unprecedented scale.
The European Commission’s consultation process relies on the principle that submissions reflect genuine public opinion. However, the deployment of AI tools like Your Voice Your Choice introduces a new dimension of manipulation, where the boundaries between authentic citizen engagement and corporate-driven advocacy blur. The tool’s design appears to prioritize alignment with PMI’s interests, even when users’ initial inputs did not reflect them. This raises questions about whether such tools should be permitted in regulatory consultations at all, given their potential to distort democratic deliberation.
Harm to Epistemic Integrity and Democratic Processes
The primary harm identified in this incident is the erosion of epistemic integrity—the trust in the accuracy and authenticity of public input. When AI-generated submissions are indistinguishable from genuine citizen responses, policymakers and the public may be misled about the true distribution of opinions. This undermines the legitimacy of regulatory decisions, as they may be based on artificially inflated opposition rather than a balanced representation of societal views.
The incident also poses a threat to democratic integrity, as public consultations are a cornerstone of participatory governance in the EU. The use of AI to manufacture consent or opposition undermines the principle that policy should reflect the will of the people, not the strategic interests of corporations.
By flooding the consultation with industry-aligned submissions, the tool may have delayed or weakened regulatory measures that could have reduced tobacco-related diseases. The OECD’s classification of this incident highlights these interconnected risks, where AI-mediated influence operations directly impact both democratic processes and public welfare.
Philip Morris’ Response and Industry Implications
Philip Morris International has denied allegations that it ‘rigged’ the EU consultation, stating that the tool was designed to facilitate public participation rather than manipulate outcomes. The company argues that users retained control over their submissions and that the AI merely assisted in drafting responses. However, journalistic investigations have contradicted this claim, showing that the tool often altered or expanded user inputs in ways that favored industry positions.
This incident is part of a broader trend where corporations and interest groups use AI to amplify their influence in regulatory processes. If unchecked, such practices could set a precedent for other industries to exploit AI in policy-making, further eroding trust in democratic institutions.
Without intervention, the risk of AI-mediated influence operations will likely grow, particularly as large language models become more sophisticated and accessible.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.