‍

Artificial intelligence is quietly reshaping the contours of our democracies, not dramatically, not with the fanfare of an announced revolution, but through successive touches, in our news feeds, in our digital polling places, in the algorithms that guide what we read, what we believe, what we get indignant about.

The question is no longer theoretical; AI is already at the heart of the political game. What remains to be seen is whether it will strengthen or hollow out the democratic project.

When election campaigns are played out behind the scenes of algorithms

During the last European elections, many voters received tailored political messages, short videos, calibrated slogans, and targeted promises, without always knowing that this content had been designed or optimized by artificial intelligence tools.

This is not a mere anecdote but a fundamental shift; the visible and collective confrontation of ideas tends to be replaced by individualized, invisible persuasion that silently influences each individual.

The French figures are staggering: according to an Ipsos study published in February 2026, nearly one in two French people has already used or plans to use generative AI to get informed about politics, especially with the municipal elections in 2026 and the presidential election in 2027 approaching.

These uses, particularly widespread among young people, raise questions about the role of AI in opinion formation, informational sovereignty, and democratic functioning.

In Romania, the first round of the presidential election in November 2024 had to be cancelled by the Constitutional Court, following accusations of foreign interference via disinformation campaigns on social media.

In Taiwan, digital manipulation operations targeting young voters were documented during the presidential election that same year. These are not isolated cases; they illustrate a fundamental trend.

Three visions of the future for the same technology

Faced with these risks, researchers and institutions do not agree on the answer. In a report published in early 2025, Professor Yann Algan and Gilles Babinet of HEC Paris identify three major visions for the democratic future in the age of AI.

The first is defensive; it aims to protect democratic institutions from the harmful effects of AI: deepfakes, emotional manipulation, and the polarization of public debate.

Proponents of this approach call for strengthened legal frameworks, independent regulation, and algorithmic transparency. Europe, with its AI Act adopted in 2024, is aligned with this approach.

The second vision is more ambitious, exploring AI's potential to improve citizen deliberation and public decision-making. Taiwan offers a concrete example here: the country has been using AI-assisted open consultations for several years to co-construct certain public policies.

From this perspective, AI can reduce cognitive biases in administration, make institutions more responsive, and give citizens a more genuine voice in complex processes, which is the meaning some attribute to the expression "augmented democracy."

The third vision is radically different and deserves serious examination, even if it raises deep reservations.

Championed by certain libertarian figures in American tech, it envisions decentralized algorithmic governance, in which the state would withdraw in favor of autonomous organizations managed by algorithms.

The question that then arises is not technical; it is political: if algorithms are driven by private interests, who guarantees their fairness? Where does democratic accountability lie?

The real problem : who leads AI and democracy initiatives ?

One of the most striking findings in the HEC Paris report is also one of the least discussed in public debate: initiatives combining artificial intelligence and democracy are currently largely driven by civil society and major technology companies.

Governments and universities remain largely on the sidelines, and these initiatives are highly concentrated geographically, primarily in the United States.

In other words, the major democratic issues of our time are being addressed primarily by private actors, often without a clear democratic mandate. This imbalance is perhaps more concerning than any deepfake.

Conclusion

For AI to serve democracy rather than undermine it, several conditions appear essential.

The first is algorithmic transparency. Citizens have the right to understand how the political content presented to them is constructed, amplified, or filtered, and citizen assemblies dedicated to AI governance could be a relevant democratic tool for exercising this oversight.

The second condition is education—not just technical AI training, but a culture of critical thinking, cooperation, and collective resilience in the face of saturated information environments. This education must begin early, long before today's children become tomorrow's voters.

The third, often overlooked, condition is addressing the root causes of democratic disillusionment. The crisis of representative democracy largely predates the rise of AI and is rooted in a sense of abandonment, in the distance between institutions and people's concrete lives. However, no technology, no matter how promising, will solve this problem if the social and political conditions that generate it are not addressed.

AI can be an extraordinary tool for deliberation, access to information, and citizen participation, but for this to happen, it must be framed by democracy, not the other way around.

‍