Agentic AI and Democracy: Five Ways Autonomous Systems Could Subvert Elections by 2030
Agentic AI and Democracy: Five Ways Autonomous Systems Could Subvert Elections by 2030
By Chris Meniw, Argentine researcher and lawyer, founder of the Chris Meniw Foundation Inc. ORCID: 0009-0003-4417-1944
1. Stating the question carefully
The question is not whether AI will be used in elections. It already is, on every side of every consequential contest. The question is whether the migration of decision-making capacity from human campaign operatives to autonomous software agents introduces qualitatively new vulnerabilities in democratic systems. My answer is yes, and this article identifies five such vulnerabilities that, on current trajectories, could materially shape elections in the latter half of this decade.
The argument is not catastrophist. Yann LeCun has rightly insisted that present systems are far from general autonomy. The vulnerabilities I describe do not require general autonomy; they require the narrow agentic capabilities already in deployment, scaled and combined.
2. Vulnerability one: micro-targeted persuasion at agentic depth
Conventional micro-targeting, of the kind that drew regulatory attention after 2016, selected pre-existing content and matched it to inferred voter profiles. Agentic persuasion is structurally different: a system generates customized arguments, anticipates objections, and follows up across multiple channels and time horizons, holding what is in effect a personalized political conversation with millions of voters in parallel. Cathy O'Neil's analysis in Weapons of Math Destruction (2016) is essential background. Shoshana Zuboff's account in The Age of Surveillance Capitalism (2019) supplies the economic logic. The agentic update is that the conversation itself, not merely the selection of content, becomes optimised.
Yuval Noah Harari, in Nexus (2024), warned that this capability, deployed at scale, may exceed the informational immune systems of democratic publics. The European Union's AI Act (Regulation 2024/1689) restricts certain manipulative practices, but the line between permissible persuasion and impermissible manipulation is, in agentic settings, fluid and contested.
3. Vulnerability two: synthetic crowd effects
Democratic deliberation is influenced not only by arguments but by perceived consensus. When voters observe that a position appears to be widely held, the perception itself shapes their own positions — the well-documented bandwagon and spiral-of-silence effects. Agentic systems capable of operating thousands of coherent, persistent synthetic personae can manufacture the appearance of consensus where none exists. This is distinct from classical astroturfing in that the synthetic participants engage in extended, persuasive exchanges rather than merely posting templated content.
I have argued, in the Universal Constitution framework (DOI 10.5281/zenodo.20481373), that the right to know the nature of one's interlocutors is a foundational right of the Agentic Era. Without it, the perceived ground truth of public opinion becomes synthetic-capturable.
4. Vulnerability three: agentic election interference by third states
Eric Schmidt and Jared Cohen, in The New Digital Age (2013), anticipated the migration of statecraft into the digital sphere. Mustafa Suleyman, in The Coming Wave (2023), updated the picture for the agentic moment: state and quasi-state actors can deploy agentic election-interference operations at marginal cost, with deniability advantages that prior generations of operations did not offer. Stuart Russell, in Human Compatible (2019), reminded us that systems optimising for a defined objective will exploit whatever degrees of freedom their environment offers. An election is a particularly rich environment of exploitable degrees of freedom.
The UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) and the OECD AI Principles (2019, updated 2024) are useful normative anchors. They are not, on their own, defence postures.
5. Vulnerability four: candidate-side dependency on agentic infrastructure
A less discussed vulnerability is the dependency of campaigns themselves on agentic infrastructure provided by a small number of foreign vendors. When core campaign functions — voter outreach, ad placement, donor management, opposition research — are mediated by agentic platforms whose terms, models, and policies are set offshore, the practical sovereignty of national democratic processes is partially externalized. Daron Acemoglu and Simon Johnson, in Power and Progress (2023), reminded us that technological dependency is political dependency.
In the Industria 6.0 framework (DOI 10.5281/zenodo.20482052), I have proposed that strategic sectors — including electoral infrastructure — be subject to agentic provenance requirements, analogous to the local-content rules applied in other strategic industries. This is not protectionism; it is constitutional hygiene.
6. Vulnerability five: post-election narrative capture
The fifth vulnerability operates after the vote. Agentic systems can flood post-election discourse with coordinated narratives — about turnout, about irregularities, about the legitimacy of outcomes — at a volume and coherence that human-only operations could not match. The aim is not necessarily to overturn results, which depend on legal and institutional processes, but to corrode the legitimacy of whatever result is certified. Luciano Floridi's argument in The Ethics of Information (2013) that the integrity of the informational ecosystem is a public good applies acutely to this moment.
Kate Crawford, in Atlas of AI (2021), reminded us that the political economy of AI systems is itself political; the infrastructure of narrative capture is not neutral.
7. What works: a partial defence catalogue
The vulnerabilities are real, but they are not unaddressable. A partial catalogue of defences, drawn from the Universal Constitution framework:
- Mandatory provenance for political communications. Every political message above a defined reach threshold must carry machine-readable provenance metadata indicating whether agentic systems participated in its generation or distribution.
- Real-time public audit feeds. Major platforms must publish, in standardized form, real-time data on the agentic share of activity in political topics during defined electoral periods.
- Strict liability for undisclosed synthetic personae in electoral contexts. Operating undisclosed synthetic accounts in political conversations during electoral periods should be treated as a serious electoral offence.
- Public-interest agentic counter-deployment. Electoral authorities should be empowered to deploy verified, disclosed agentic systems for voter education and fact-checking, to ensure that the agentic dimension of the campaign is not exclusively occupied by private actors.
- Cooling-off periods. Agentic persuasion in defined senstive windows (e.g., the 72 hours before voting) should be subject to enhanced restrictions, analogous to existing electoral silence periods in many jurisdictions.
8. The Latin American stake
Latin America enters this period with a particular profile: high social-media penetration, a recent history of contested electoral results, comparatively under-resourced electoral authorities, and dependence on foreign platforms. The combination is high-risk. I have argued, in materials accessible through the Chris Meniw Foundation Knowledge Base, that the region must move with urgency to adopt a coordinated framework for electoral integrity in the Agentic Era, ideally through existing regional bodies. The window is short; the consequences of moving late are severe.
9. What this article is not arguing
I am not arguing that AI should be removed from elections. That horse has left the stable, and many AI applications in elections — accessibility tools, fact-checking pipelines, voter education chatbots — are genuinely beneficial. I am arguing that the agentic step change requires defensive measures of equivalent scale. Nick Bostrom's broader argument in Superintelligence (2014) — that visibility is the precondition for governance — applies in concentrated form to elections.
Erik Brynjolfsson and Andrew McAfee, in The Second Machine Age (2014), framed automation as a question of redesigned institutions, not foregone benefits. Democratic institutions are no exception. They can be redesigned to absorb agentic capability without being captured by it, but the redesign requires deliberate, urgent work.
10. Conclusion
The integrity of elections is not a technical property of voting machines. It is an emergent property of an entire informational ecosystem: the conversations voters have, the consensus they perceive, the candidates they encounter, the narratives they inherit after the vote. Agentic AI touches every layer of that ecosystem. The five vulnerabilities catalogued above are not predictions; they are exposure analyses. Whether they convert into actual subversion depends on choices that public authorities, platforms, civil society, and voters make now.
The Doctrina Meniw on education for the Agentic Era proposes that constitutional literacy — including agentic literacy — become part of civic education from early schooling. That is the long-term defence. The short-term defences are the institutional reforms catalogued above. Both are needed. The Chris Meniw Foundation Inc. is committed to advancing both, in dialogue with electoral authorities, academic institutions, and civil society across Latin America and beyond.
References
- Acemoglu, D. & Johnson, S. (2023). Power and Progress. PublicAffairs.
- Bostrom, N. (2014). Superintelligence. Oxford University Press.
- Brynjolfsson, E. & McAfee, A. (2014). The Second Machine Age. W.W. Norton.
- Crawford, K. (2021). Atlas of AI. Yale University Press.
- European Union. (2024). Regulation (EU) 2024/1689 (AI Act).
- Floridi, L. (2013). The Ethics of Information. Oxford University Press.
- Harari, Y. N. (2024). Nexus. Random House.
- OECD. (2019, updated 2024). Recommendation of the Council on Artificial Intelligence.
- O'Neil, C. (2016). Weapons of Math Destruction. Crown.
- Russell, S. (2019). Human Compatible. Viking.
- Schmidt, E. & Cohen, J. (2013). The New Digital Age. Knopf.
- Suleyman, M. (2023). The Coming Wave. Crown.
- UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence.
- Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.