Committee on Civil Liberties, Justice and Home Affairs Joint meeting Committee on Constitutional Affairs Special Committee on the European Democracy Shield
Legend
Transcript markers
- text Sentence being played
- text Search result
- text Intervention cited by the link you opened
Click a sentence to play the video from that moment. Select text to quote it with the speaker, timestamp and link.
More options
LIBE European Parliament
Follow its sittingsJavier Zarzalejos
Natalie. Welcome. The chair of the UDS special committee. The chairs of the INCO and AFET committee have also been invited. The public hearing will include interventions by three guest speakers, and aims to address topical aspects within the areas of competence of our respective committees.
As regards the Lieber Committee, the speakers will focus on fundamental rights and AI-driven elections. AI driven tools are increasingly shaping the way in which citizens access information, form political opinions, and participate in public debate, while these technologies can offer new opportunities for engagement and access to information. They also raise significant challenges with regard to the protection of fundamental rights, including privacy, data protection, and freedom of expression. Among the key concerns are the risks of manipulating through AI generation, disinformation and deepfakes, profiling and the micro-targeting of voters, the use of algorithms to influence the visibility and dissemination of political content. The potentially undermining of electoral integrity and citizens' trust in democratic institutions.
European Union has already taken important steps to address these challenges through legislative initiatives and efforts to increase transparency in political advertising, ensuring that effective implementation and continuing to assess the impact of emerging technologies on democratic processes will remain crucial. Today's hearing offers an important opportunity to exchange with our guests on how elections are evolving in the age of AI, and reflect on the safeguards needed to strengthen our democratic resilience. We have a very tight agenda. Therefore, I kindly ask all the participants to respect the timing allocated to them. And now we'll pass the floor to Missus Natalie Loisso, chair of the special committee on the European democracy shield, for her opening remarks.
Nathalie Loiseau
Madame Blaseau. Thank you, chair.
Colleagues, guests, our discussion today addresses, as you have just said, a profoundly transforming challenge across democracies worldwide, namely how to safeguard the resilience of our electoral processes in the face of artificial intelligence. Of particular concern, the scale, speed, and sophistication that artificial intelligence enables at a fraction of the cost of what was possible before. The examples of this are numerous. Looking at the potential for artificial amplification, we remember the presidential elections in Romania, where the French agency, Vigi Noum, documented how coordinated networks exploited TikTok's recommendation system. More than 25,000 coordinated accounts generated more than 73,000,000 views for a single political hashtag in just one week.
On deepfakes, in the last elections in Moldova, in just 48 hours, a fake video was viewed over 350,350 times on TikTok and Telegram, which is a significant number for a country with a voting population of roughly 2,500,000. Talking about deepfakes. In my country, France, we have local elections at the moment. One party, one candidate was the target of a deepfake generated by artificial intelligence and a fake campaign website. It's another example.
Before the Dutch parliamentary elections, the data protection authority tested widely used AI chatbots, including ChatGPT, Gemini, and Grok. The results showed that these systems provided distorted and polarizing voter voting advice to a significant number of voters, which influenced their voting decisions. With over 56% of recommendations directing users towards just two political parties, while many other parties were rarely mentioned at all. We know that AI also offers opportunities for strengthening electoral integrity and for bolstering cybersecurity by detecting in real time manipulation and interference. So AI can help us master the very challenges it creates.
So my hope today is that we, in our discussion, will look at ways in which we can ensure the right balance in our approach.
Javier Zarzalejos
Madame Blassault. And now I give to Mister Charles Gerens, vice chair of the committee on constitutional affairs, for his introductory remarks.
Charles Goerens
Thank you. Dear colleagues, dear guests, I think we can all agree that artificial intelligence is increasingly affecting not only political communication, but also the broader electoral environment and the functioning of democratic institutions. As new technologies reduce the cost and speed at which interference can occur, they raise questions about the resilience of electoral procedures, the preparedness of electoral authorities, and the safeguards needed to preserve trust in democratic processes. In this context, AFCO committee has selected the expert, Missus Ines Narciso, head of information integrity and resilience, Cyberspace Cyber Peace Institution in Lisbon, who will address how AI is beginning to affect the organization and administration of elections, whether existing legal and institutional frameworks in the EU and its member states are equipped to respond to these developments, and what institutional safeguards may be required to ensure that AI strengthens rather than undermines democratic electoral systems. As we know, the EU has already established an important legal framework to address these challenges.
This includes the artificial intelligence act, which introduces the world's first comprehensive regulatory framework for AI based on a risk-based approach, aiming to ensure trustworthy and human-centric AI while protecting fundamental rights and democratic values. These issues are going to be further elaborated by our experts. But to avoid stepping over Libase and Imco's competence from the perspective of the Committee on Constitutional Affairs, the key question is complementary but distinct. How these technological developments affect democratic institutions themselves, including the organizations of elections, the functioning of electoral systems, and the preservation of democratic legitimacy and trust in electoral processes. Today's hearing, therefore, aims to bring together these perspectives in order to examine how the EU and its member states can ensure that artificial intelligence supports, rather than undermines, democratic elections.
Before giving the floor to our experts on the panel, I would first like to give the floor briefly to Mister Emmanuel Kefal Logiannis, reporter of the own-initiative report, institutional aspects of artificial intelligence in the context of European integration, and the author of the idea of this hearing. Mister Kevaloriannis has been leading the European Parliament's work in AFCO on the institutional implications of artificial intelligence for the European Union, including questions relating to democratic governance, institutional resilience, and the functioning of our democratic systems in the AI area. Mister Keefalo Yannis, you have the floor for a short introductory intervention.
Emmanouil Kefalogiannis
Thanks, chair. Dear colleagues, I would like to thank the chairs for the organization and remarks of today's debate. AI is increasingly affecting democratic processes, public debate, and the integrity of elections in the European Union. AI is shaping the way information circulates, the way public opinion is formatted, and is affecting the way democracy itself operates. For this reason, AI is an institutional and constitutional issue of the European Union.
For the Committee of Constitutional Affairs, the key question is one: how do we ensure that AI strengthens rather than weakens democratic auto accountability, transparency, and electoral integrity. The Artificial Intelligence Act is an important step for the European Union, but the rapid development of AI systems raises broader questions. AI already affects access to information, shaping public opinion, and the credibility of democratic debate. With AI, new risks emerge, deep fakes, bot networks, digital influence operations. Protected elections in a digital era are a matter of resilience.
Our response must be guided by three principles. First, the governance of AI must remain in full alliance with the values of the European Union in accordance with article two of the treaties of the European Union. AI must remain human-centered, transparent, and accountable. Second, we must strengthen parliamentary scrutiny. The European Parliament has a specific responsibility to ensure democratic accountability and institutional balance in accordance with Article 10, 13, and 14 of the treaties of the European Union.
Therefore, an observatory would help the European Union and the European Parliament to understand the institutional and democratic implications of AI on fundamental rights. Finally, the governance of AI must be addressed at a global level. AI requires stronger international cooperation. It is a systemic technology with geopolitical implications. Dear colleagues, democracy cannot function where citizens no longer know whether what they see is real, what they hear is authentic, and whether political messages are produced by humans or by algorithms.
AI can strengthen democracy, but it can also be used to manipulate it. Our task is therefore clear. In Europe, technology serves democracy. Thank you for your attention.
Javier Zarzalejos
Thank you, Mr. Gefalo Guiannis. And now we will move to the presentations by our guest speakers. I do thank you for joining us today in this public hearing and for having the benefit of your input to these three committees convened here. So, I would start by giving the floor to Missus Ines Nartizo, who's the head of information integrity and resilience at the Cyberpeace Institute and PhD researcher at the ISCTE, the University Institute of Lisbon. Your presentation, for a maximum of 5 minutes.
Miss Nartizo, the floor is yours.
Speaker 16
Thank you. I will very much try to keep a note of the time, although I do feel that it is a topic that it's hard to talk so briefly. So I will start by saying something I always tell my students, which is we should look at AI as a child with superpowers. And if we feel that we wouldn't let our children, even if they had superpowers, super strength, super intelligence, super fast, do something on their own, then we probably shouldn't let AI do it on their own as well, and there should be a human in the loop. So the first lesson, I would like to bring here is the idea that we should think that AI's interference or effects on election and electoral procedures goes way beyond the political campaign.
Speaker 21
So we need to remind ourselves that it's not just about, you know, deep fakes and about what one candidate said that was misleading about the other one. We're talking about, for example, a cyber attack enabled by AI on a critical infrastructure in the week before the elections, so that nobody has public transport and everyone gets really angry at the government. So we need to think about it as way more than the political discourse, and we need to think that it will be used by threat actors to interfere with elections in several ways and in several procedures. From voting registration, from making it harder, for example, on immigrants through AI bias to get nationality, the examples can go on. The second thing that I feel is most important to say is that, and it has been said here by Natalie, is that AI changes the opportunity structure.
Speaker 16
It industrializes opportunity, and it makes it accessible for previously very smaller threat actors to actually interfere in elections because it's become just very cheap, very fast, and possible to anyone. So that means that a candidate that is in tenth place in the polls, if he's really motivated and he wants to devote himself to interfere with elections using AI, that is now possible. So, of course, we always look at the usual actors, big threat countries that have a lot of money and resources. But right now, we're going to start seeing threats from very small actors, maybe lobbies, maybe economic sectors, maybe just disgruntled people. As us, looking at procedural and what exists in terms of AI and in terms of elections and in terms of legislation, also including the DSA and the code of practice, the code of conduct right now, I don't feel there is a lack of rules and there is a lack of regulation per se.
The main problem that we are seeing as researchers is that there is no operation or there's limited operational capacity. And that means, very specifically, that there is lack of data. So civil society and governments and law enforcement are not working effectively enough to really be able to gather the data that enables to act against situations where incidents occur. So we need more interoperability. We need guidance from the EU saying this is the framework that we're going to use.
Speaker 17
It doesn't mean it's perfect. It doesn't mean that everyone will agree on it. But unless we start collecting data in an interoperable manner to use AI on our side to understand the patterns, to understand the trends, to prevent things before they start happening. I'm not sure if it's possible to go to the slide before, because I think I jumped one inadvertently. I'm sorry.
Speaker 16
Yes. And we also need to regulate all manipulation vectors. So one thing we've been seeing is there is a lot of investment in everything that has to do with content manipulation. You know, deep fakes, a lot of the discussion on AI in elections is about deep fakes and false information, manipulated images. But actually, the threat comes way more from amplification and suppression.
And I was happy to hear also some of you mentioned that here this afternoon. Because indeed, what AI will allow is it will allow, for example, an amplification of one candidate's voice over the others. And more importantly, it will suppress voices. This means that, for example, certain targeted communities, more vulnerable communities may be suppressed using AI with hate speech, harassment, doxing, spamming, DDoS attacks against their websites, etc. I think we also need to look at how AI is gonna start integrating with wearables and neurotechnology.
So thinking about, you know, things like the idea that AI glasses will know that someone has an issue with immigrants because they cross the road when they see an immigrant, and therefore, they can target ads or targeted political content that is disinformation about immigrants to them, and how that is gonna become more and more an issue. And then I think I'm managing to get through in time. As a positive note, there is a lot that we can do, and there is a lot that IEI also allows us to do. One of them is really trying to create a regulation system that fosters and tries to make authenticity become more regulated. There are some initiatives, the C2PA initiative, for example, for media authenticity that some platforms are using, some are not, for example.
And the relevance of knowing the provenance of content is becoming more important. The second one is red teaming and this idea of stress testing that we really need to stress test systems, procedures, and really try to anticipate what can be the next thing. So getting teams on board before elections happen, and say, if you wanted to use AI to tamper with these elections, what would you do? What are the vulnerabilities that you spot? We are here today talking about AI, when AI is already an issue, and it would be interesting to start thinking about things like quantum computing, for example.
So anticipating technology and the problem it's going to bring before it is here. And lastly, but not less importantly, we need coordination. We need observatories or institutions that start collecting benchmarking data on, for example, AI. How does AI respond to a question like, who should I vote for? To see if these systems are being manipulated and to see how modules progress and to try to regulate at least based on data and evidence of real testing with different modules.
Javier Zarzalejos
Thank you so much for your attention.
Thank you. Thank you very much, Missus Narciso. We are all aware that sometimes it's difficult to pack all your ideas and what you would like to say into the time allocated, but otherwise, it would be impossible to manage these sessions. And now I give the floor to the next speaker at this point, Missus Julian Mueller, who is an associate program officer of digitalization and democracy program at the International Institute for Democracy and Electoral Assistance. Again, you have four minutes for your presentation.
Floor is yours, and thank you, Mrs. Miller.
Speaker 20
Yeah. Thank you very much for inviting me today and giving me the opportunity to speak here, and I hope I can also stick to the time. But, of course, there's so much more to say. At International Idea, we work really closely with the electoral authorities from all over the world. And so today, I want to precisely share that perspective with you.
And my message that I want to bring across here is simple. I think that AI's impact is much more long-term than we think. It's not necessarily the one single electoral defect that we need to worry about, but instead, it's much bigger. It's the slow erosion of information integrity, and also with that, our democracy that we should worry about. And I want to get into this based on four challenges that we identified.
So the first challenge is the collapse of our information system, because the key difference is not simply that there is now more disinformation. What changes is, the operational problems also that AI creates for electoral authorities. Because on the response side, electoral authorities, but, of course, also civil society, political parties, and journalists have to spend much more time to trace and verify information while attackers spend way less time producing because they need less resources and time to do so. And so it also becomes much easier, in turn, to exploit vulnerabilities in our democracies. So, very bluntly speaking, AI turns safeguarding election integrity into a firefighting exercise. And what happens when institutions are constantly firefighting, then they have way less capacity to govern the process well.
So this is not simply a fact-checking problem like we sometimes frame it. It's much bigger. It's actually a resilience problem because the question that we should ask and find an answer to is how do we protect the information space from being flooded with cheap, repetitive, and often misleading synthetic content that gradually erodes public trust and clarity. And then the next challenge is algorithmic amplification and monetization. We often focus on AI-generated disinformation itself, but what we often overlook is how AI-based algorithms shape its impact and its reach of this disinformation.
And a lot of business models of platforms reward the most engaging content, as we all know, so which is content that is most polemic, emotionally provocative and sensational. And, of course, it is also something that malicious domestic and foreign actors exploit to push their narratives. And another thing that is often overlooked is monetization. Not all actors necessarily want to attack democracy, but a lot of them just do it for the money because it pays. So even though the intention might be different, though, the consequences are the same, and it's still this kind of disinformation that still pollutes our democratic discourse.
And then the third challenge is that AI can exclude because the threats of AI do not affect all political actors equally and in the same scope and severity. And yet a lot of AI policy frameworks around the world, but also in the EU, do not adequately address how weaponized generative AI disproportionately targets female political candidates and individuals from LGBTQAI+ communities and how this also discourages their political participation. And then the fourth challenge is how AI is changing election administration itself. From our research that we conducted on an idea, we learned from a lot of electoral authorities around the world that they are using a lot of AI systems already for voter registration and candidate registration, but also for external and internal institutional chatbots. And this can bring benefits, but of course, also new risks.
And here, I want to highlight one gap of the EU AI Act because these kinds of AI systems or Annex III classifies those AI systems as high risk that are intended to influence an election, the outcome of an election, or voter behavior. But, of course, a lot of these AI tools are not inherently designed to do so, but by consequence, because any AI system can have inherent bias or hallucination or error, it can have the same consequence. So this is something I think we need clarification on, and to make sure that also such scenarios are covered by regulatory safeguards. And so I want to sum up by saying that in 2024, at IDEA, we suggested it's not really necessary to panic over AI. But perhaps now it might be time to panic because we are really running the risk of a collapse of our digital public square, because in a society that is unable to have a shared reality and to discuss its problem, then is that still a democratic society?
So what we should do is that we should build on what we already have in the EU. We should close gaps in our EU digital rule book to better oversee AI's impact on electoral processes, specifically. And another thing that we suggest is that, for example, the digital service coordinators or also the AI office should include more external election or should include more electoral expertise to also translate better DSA and II Act obligations into the realities of electoral periods. And then third, also, the European Centre for Democratic Resilience should be more operational to adequately respond to electoral risks. And that should also include structured consultative measures to, for example, include electoral authorities and support or better support their work and realities.
And then also, we can leverage the existing networks like the European Cooperation Network on Elections, which already serves as a very strong vehicle for improving election authorities' ability to understand, assess and respond to AI related risks. But of course, this work should be supported through structured cooperations with organizations and experts already engaged in research and technical assistance, and to make sure that we have the necessary resources and capacity to do so. Thank you very much.
Javier Zarzalejos
Thank you very much for your presentation, Miss Miller. And now I will give the floor to our third and last speaker, Miss Sofia Calabresse, who's a digital policy manager at the European Partnership for Democracy, again, for five minutes.
Speaker 22
Thank you very much. Good afternoon, everyone. Thank you for giving me the opportunity to take the floor today. This discussion is very important. As mentioned, I work for the European Partnership for Democracy.
We are an umbrella organization. We have over 20 members in different European countries who work on several aspects of democracy support, from more traditional issues linked to democracy, such as rule of law, to new issues such as the impact of new technologies on democracy and elections. That's also my main field of expertise. I will try to integrate what was said by my colleagues. I work in particular at the EU level and on the EU regulatory framework, so that's what I'm going to focus on in my short intervention today.
I think the main threats were, like, already covered very well by my colleagues. I only want to make three points regarding the EU regulatory framework. The first point would be that we actually have a very advanced regulatory framework in terms of digital policy, in terms of AI regulation, online platforms regulation, data protection. Our digital framework was not only thought to regulate companies and the business sector, but it was also already drafted having in mind our fundamental rights and EU values and the protection of democracy. Right now, it is tied up even better with the European democracy shield.
But if we look at the rules that we have, I think the issue that we have right now is not really about adding some additional rules or regulations, but it's really about implementing and enforcing boldly what we have. We shouldn't be ashamed of our digital framework. We should actually be proud of it, and we should implement it strongly. That said, obviously, as a second point that I would like to make, there are always some adjustments that can be made, and we are still learning how to implement this new regulation because in most cases, it's very recent pieces of legislation. Even the ones that have been applicable for a while, such as the Digital Services Act, it's still, in most cases, a learning process, in understanding how to best utilize the potential of the regulation.
So, for example, when it comes to the Digital Services Act, I think there are aspects that can be improved, especially when it comes to the assessment of risks by very large online platforms. A lot of the issues that were mentioned linked to elections and civic discourse online should be part of this risk assessment and can be considered by platforms in a more detailed way, in particular when it comes to AI-generated content, but also political ads campaigns and disinformation campaigns, but also about the structure itself of these platforms. So the recommender systems and their monetization business models that were mentioned before by some colleagues as well. When it comes to some of the other rules, there are similar observations that can be made. I think some gaps in the AI act were already mentioned when it comes to the risk assessments for AI systems intended to be used to influence elections.
That intentionality is actually very restrictive. So this would benefit from some clarifications or a broader interpretation. And another tool that we have at our disposal is obviously the political ads regulation. The political ads regulation also presents a few technical issues here and there, but I think one of the most important aspects that needs to be noted is that, because a lot of online platforms have announced bans on political ads on their platforms. Some rules are considered less effective, but the thing is whether or not the platforms have internal policies that consider bans, doesn't mean that certain content cannot be on the platform, such as in the case of TikTok, which on paper doesn't allow political ads.
So we should actually enforce these rules, even when on paper, the platform says that they don't allow certain content because the laws apply when there is, like, the specific case that is in the scope of the law. And that also applies to political influences, that it's more difficult to catch within the scope of the law, but they should also comply with the transparency obligations that are foreseen. And I think a last point that I wanted to make is broader. It's linked to these values that we have been talking about, and about the fact that we should integrate these values into the technologies we use. I think we should look at that also in a more constructive way.
We're discussing a lot in the EU about fostering the EU tech sector. This is obviously a very ambitious objective. But when we do, we should actually create a tech sector that is different from what we have right now. And this can be done under different aspects. There is obviously a regulatory aspect.
There should be compliance with the rules that we have. And as I pointed out, the rules are quite strong, so we should use them. There is a more technical perspective. I think a few elements were mentioned about interoperability and about open source software to be used in alternative models to be put forward. And then there is one last point, that is the organizational model.
Obviously, we cannot really expect from companies that are very top down, very often not based in Europe, and have an exclusively for profits goal to actually make the interest of EU citizens. So I think we should look also into alternative models in terms of ownership, purpose, and control, when building a new tech sector. And alternative models, I'm talking about, for example, having more hybrid models between private and public. So public-private partnerships, cooperative models, foundation-led. I think I will stop here.
The three main points were let's use our regulation because it's good with some adjustments, and let's look at the broader picture, and let's be constructive as well. Thank you very much.
Javier Zarzalejos
Thank you. I thank you for your three very substantial presentations. You really make the most of your time. I'm sure that it would be the basis for a very productive debate. And also, you will have the opportunity to touch upon other aspects of this matter in the course of the Q and A session that will be organized in two rounds.
Mister Garends, the floor is yours now.
Charles Goerens
Thank you, dear colleague. During the first round, we will first take questions of Liebe, Afco, or EUDS members for two minutes each, starting with the group EPP, followed by S&D, PPE and ECR. Our guest speakers will then have eight minutes altogether to reply to the questions. As instructed in advance, the groups can decide to allocate the speaking time to one member for two minutes or divide it between two members for one minute each. I will start by giving the floor to Mister Cunha from the EPP for two minutes.
Thank.
Paulo Cunha
you. Artificial intelligence in the way that we communicate and the way that we receive our information in public life is going to go through some unprecedented changes. And today, I think we're seeing misinformation or disinformation campaigns, which are personalized, automatized, and very often amplify impressions without the average citizen being aware of that. The European Union has tried to respond to this with different pieces of legislation. But there is still a key idea here that technology should serve democracy rather than the other way around.
And, yeah, I think we have clear political responsibilities. We need more transparency, and we need to make sure that democratic principles meet the demands of a world in which technology is progressing very quickly. I have a very specific question for Missus Narciso. If the next generation of AI, which threatens our democracy, is algorithmic and transnational and we can't find out who's behind it, how will the European Union deal with that? I. E, shifting to a preventive operational model in order to anticipate and neutralize threats in real time?
Because I think the question here is quite simple. If we don't protect our elections, then we will undermine the very heart of European democracy.
Charles Goerens
Thank you very much. Two S and D speakers on my list, Juan Fernando Lopez Aguilar for one minute, followed by Mrs. Birgit Sipple for one minute as well.
Juan Fernando López Aguilar
Certainly, it won't be the end of it. We are witnessing a technological revolution, which is unprecedented in the history of mankind. But what brings us together, members of the Committee of Liberty, Justice and Home Affairs, and members of the Constitutional Affairs Committee, is its impact on democracy, impact on law, impact on fundamental rights. That's our concern because it's only obvious that in this revolution in which it will become impossible to tell the difference between fact and fake, between truth and lie, everything that is relevant for us as the European parliament is at stake. So I got the question.
First, as to the platform's tools to identify risks in artificial intelligence-driven disinformational deepfakes, is there anything better that we could do? Is there anything that could be improved to fill the gap between formal compliance and its real impact on fundamental rights, rule of law, and democracy, and to protect the integrity of electoral processes? Most importantly, that's my minute. Thank.
Okay.
Charles Goerens
Now I'm going to give the floor to the PFE representative, mister Laszlo.
You. I will speak again. Two.
Minutes.
András László
The European Commission, this very week, acted upon its censorship law based on the DSA in the run-up to the Hungarian elections. I'd like to remind you that national elections are part of national sovereignty and within the competence of member states, and they are not the competence of the EU. So my question to experts is, who is responsible? Who's taking responsibility when you are saying, as partners of the European Commission, that there is likely to be some kind of interference or intervention in election and something needs to be done? Because what we're seeing is that the censorship partners of the commission are all radical left-wing media organisms or organizations, which were financed by the commission by USAID or National Endowment of Democracy.
And the Trump government already identifying them as some kind of activist left wing center or network, and the European Union is still supporting these organizations. We are not going to let you do the same thing like you did after the Romanian elections. The judicial committee of the American House of Representatives clearly proved that the European Commission has interfered with a number of national elections in Europe, as well as the European parliamentary elections in 2024. Who is going to take responsibility for this? Please tell me.
And who is taking responsibility for the manipulation or for saying what is AI-generated manipulation and what is not. MEPs and members of parliament have the responsibility to regulate public debate and to regulate national elections. So this is up to national MPs and not the European Union. Thank.
Charles Goerens
you. And the last speaker for the first round is from the ECR, George Pipera.
Thank you. I'll speak Romanian.
Gheorghe Piperea
We are talking about an instrument that is omnipresent. It is ever present on the economic stage and in the public square, including when it comes to elections. And I speak as a citizen of a member state, that has gone through a traumatic experience, the cancellation of our elections, an experience that no one should ever have to live through again, which unfortunately could become a reality again, unfortunately, because AI is to be involved in elections again. As Noam Chomsky said, there are systems whereby consent can be fabricated, including consent when we make elections and electoral choices that are based on ideology, especially if we are to leave AI in charge of these things. AI already raises issues pertaining to the rule of law, constitutional issues, and issues that have to do with rights and liberties.
And the major thing that we contemplate as elected representatives of millions of European citizens is that there is no person who can take responsibility for what AI can do, for what could go wrong if AI becomes involved in election processes. There are also problems that arise from the fact that AI is becoming integrated in the fundamental values that the EU is based on. It is extremely important that we do not forget about democratic legitimacy, that we do not forget about our responsibility toward those who put us here, people who elected us. We should not allow AI to draft laws, case laws, and go to the ballet turns instead of us. Let us not allow Oculus-type smart glasses to distinguish between Europeans and migrants.
We should remain capable of distinguishing between a person who is manipulated by AI and a person who has the freedom to make a free choice. From the Imco perspective, it is extremely important to note that artificial intelligence is a tool that can go wrong. It is an ideologically manipulated instrument that has cognitive bias. At the moment, there is no way to protect ourselves from its hallucinations or lies. When we talk about AI from the point of view of the rule of law and democracy, we need to, first and foremost, not forget that we can go wrong.
Charles Goerens
That takes us to the time for answers from our experts. You have eight minutes. Go ahead, please, Missus Calabresi.
Speaker 22
Start because there was a very specific question on the PSA risk assessments that I think it's important to respond to. So it is true that the tool in the DSA of the risk assessments was something that was put forward to sort of, like, move somehow the burden of proving certain risks, and mitigate them to the companies because to the platforms because they have more access to data to identify specific risks and put forward mitigation measures. So as sort of an initial exercise, to better understand, what are the main threats and how to mitigate them? This exercise has been ongoing for a couple of years. It relies a lot on the actual goodwill from the platforms.
And so far, actually, at EPD with our colleagues from Liberties, which is another democracy support organization, we have conducted analysis of these risk assessments in the past couple of years when it comes to civic discourse or electoral processes. And so far, they have fallen short. So this was the concern that was raised, that platforms are only ensuring compliance in a very formal way. But content-wise, there is not that much that we can draw from this risk assessment. So, I think what can we do about that?
On one hand, the European Commission should, like, push more strongly to actually have stronger risk assessments, with better identification of risks. A lot of our colleagues from different civil society organizations have very well identified a range of risks that are not present in this risk assessment. Same for the mitigation measures. If we can do that, they can do it as well. So first, the feedback to the platforms should be stronger.
And if they don't comply, then, obviously, there will be legal action as well. There should be legal action as well. But then I think the other step is really, and this was mentioned before by some of my colleagues, is really the access to data is crucial because there are issues that from an external perspective, we still don't fully understand about the functioning of, like, these platforms and about the main risks that they put forward. So, actually, unlock the potential of the access to data from the DSA, that could be very important so that a researcher can also do that sort of exercise as well.
Speaker 16
So we are here today to talk about artificial intelligence. And despite the discussion that we may have and which is, of course, political on removing content and what is, you know, the truth and the veracity of content. I think we need regulation, and legislation should start on the things we all agree on and we have consensus on. And one of them is that we don't want machines deciding for us, and we don't want machines speaking for us. When we look about interference and manipulation into elections, we are talking about amplification and suppression that have, for example, nothing to do with the content being true or false.
We are talking about the same reason why candidates have the same media time during television interventions. Why should a platform allow for one candidate to artificially buy machines, have all of the space, which was the example that we gave, for the Romania-like elections. So to regulate on coordinate coordinated inauthentic behavior, we need more data, and we need design changes in platforms. I'm from a university that was always very much against the removal of content, and more about giving context about content. Meta tags, for example, for identifying AI content, are failing.
Yesterday, indicator released a research in which they proved that all platforms, especially Instagram, are failing to identify clearly AI-created content when uploading. So, the AI meta tags are not working. So, these are all small things that we can do that are doable, practical, and operational. So we need to become more operational in the way that we implement regulations, and that means talking with the experts who are on the ground handling things directly.
Speaker 20
Yes. I want to answer the question on why this is also an EU matter and not just a national concern. Because I think when it comes to interference by foreign or also domestic actors, it's neither one or the other. I think it's both. Right?
I think we need still, of course, national strategies and not taking away, of course, responsibility from national election authorities or national regulators. But, of course, we also need an EU-wide strategy because if all 27 member states have their own strategy on FIMI, then malicious actors can exploit vulnerabilities of their member state with the weakest protections, and it can start there, but then filter into the other member states as well. Also, because the nature essentially of our digital public sphere now of online platforms is cross border, because a content can be published or created in one member state. It can be posted in another member state and essentially can affect or have an effect on elections in a third member state. So it's already a very changing, dynamic in the where the threat takes place.
And so that requires, for sure, EU wide strategy. And instead, we should also use EU institutions, EU mechanisms, to help support member states in tackling these challenges because, for example, some electoral bodies of some member states may have less capacity to identify risks or to have enough capacity to mitigate these risks. So they can draw on EU support mechanisms to help them with those challenges, but also essentially for electoral authorities to exchange and to use EU platforms to understand and learn from each other. And finally, also because we are using EU relationships can be very useful, especially when we think of partnerships with large online platforms. So because the EU, as a big economic hub, has so much more leverage than one individual member state.
So that's why we need a strategy as an EU to build those partnerships with platforms, but then member states can draw on that. So it's reinforcing each other. It's not one or the other. We need both. So this is why I think it's not only a national concern.
Thank you.
Just.
Charles Goerens
Before we move on to the, or I give the floor to the next chair, I want to say that I have given my minute to Mrs. Natalie Loiseau.
Nathalie Loiseau
Yes. Thank you, Chair. The second part, we'll continue with questions from Libya Afco or EUDS Maps for two minutes each from Renew Greens, the Left and ESN. Guest speaker will then have, again, eight minutes maximum to answer the MEPs' questions. As in the first round, the groups can decide to allocate the speaking time to one member for two minutes, or divide it between two members for one minute each.
And we'll start with renew, and thank you very much, Charles. And.
I'm going to shift to French to ask a couple of questions, specifically for Missus. At the moment, an omnibus package is being negotiated. So the digital omnibus package is intended to simplify EU regulation or make them more consistent in this area. And I wonder to what extent those who are working on this certification are getting the alarm bells or understanding the alarm bells that you are sounding. The simplification, I'm not sure it's gonna have an added effect on the legislation we've set up, and whether it takes into account or the simplification takes into account the risks arising from a generalization of the use of AI.
I also wonder to what extent the different commission directors general have started to put democratic issues at the heart of what they're doing, by which I mean, if you speak to the DG Connect, well, they talk about digital innovation. DG Just will tell you about the implementation of the current legislation. And the center for the metric democratic resilience is going to shift to DG Com. Now I wonder in your work that you carry out if you get the impression that these DGs actually speak to each other. Because when it comes to protecting democracy and elections, I wonder if it's fully taken into account by all of those involved.
Other members. And for the Greens, the floor goes to Leoluca Orlando.
Thank.
Leoluca Orlando
Thank you very much. I'll be very brief. There's the institutional report on artificial intelligence as part of the European integration. And here, there are two key points: sovereignty and trust. And I think we need to look at this and understand that there is a lot of mistrust of artificial intelligence, and so we need a proper political decision to build up trust.
And then, at the same time, we need to make sure we have European digital sovereignty. And remember that there are major powerful private groups outside the European Union which need to be tackled. Can we do this? I hope we can. I will give the rest of my minute to Mrs.
Strecker.
Tineke Strik
Thank you very much, dear three chairs, and dear three speakers. I have one question for Mrs. Calabrese, and this is about, you mentioned the need for shared benchmarks within the EU. Maybe you could elaborate a little bit on how they would look? And also, how could we better anticipate the use of AI in the elections?
And the other question is for Missus Miller, because the Hungarian elections were already referred to. It will take place within less than four weeks. And, we have very worrying news about Russian intelligence interference, that is about to emerge, where we see that the National Election Commission decided not to interfere, neither the Hungarian intelligence services. So this is, of course, a problem in a captured state, where there is a sort of deliberate cooperation with foreign interference actors. And my question is actually, in these kinds of situations, what could the EU do?
We have this European Centre for Democratic Resilience. Could they do more than simply monitor? Could they also put more pressure on the Hungarian authorities to stop this and to interfere in this interference? And what could the European Commission do if it comes to the enforcement of our legislation and our values? Thank.
Nathalie Loiseau
you. I see no speaker on the left. So I give the floor to ESN, Christine Anderson.
Christine Anderson
Yes. Thank you. This hearing addresses democracy and elections in the AI era. AI changes the scale, the speed, and the tools of communication, but we should be very careful not to equate the use of AI with abuse. The existence of misuse does not justify a shift towards managing the overall information environment.
We should also remember that AI can amplify something entirely legitimate, citizens organizing, persuading, contesting ideas, even mainstream ones. Every democratic movement has always relied on amplification from pamphlets, leaflets, to modern digital tools. AI does not change the nature of this activity. It changes its efficiency and its reach. Automated tools, including bots, can function as the digital equivalent of distributing leaflets at scale.
They replicate and disseminate messages. Can it be misused? Of course. But amplification as such is not illegitimate. So if we begin to treat scale, repetition, or automate automation itself as suspicious, we are in essence classifying normal democratic activity as a threat.
And this brings me to a second point. When AI systems curate, rank, or prioritize content, they are effectively making editorial decisions. At the current stage of AI development, these decisions are always traceable to a human actor who has designed, instructed, or deployed the system. In that sense, AI is not replacing editorial freedom. It is an instrument through which it is exercised.
This is, in essence, an extension of freedom of expression and, where applicable, the freedom of the press. So the key issue is not whether AI creates risks, it does. The key issue is whether we remain focused on illegal conduct or whether we begin to govern lawful, democratic discourse. And on that, I would like to ask three questions to the experts. First, how do you clearly distinguish between illegal AI-enabled manipulation and lawful political speech that is simply amplified or controversial?
Second, who ultimately defines harmful narratives or information integrity, and what democratic or legal accountability exists?
Can you?
Conclude, please?
Christine Anderson
Decisions? Third, when AI systems make editorial choices at scale, do you consider this an exercise of human editorial freedom? And if not, what legal basis do you have to restrict it?
Nathalie Loiseau
Thank you. Thank you. Now, I give the floor back to our experts for replies, starting with Miss Calabrese.
Speaker 22
Yes. Thank you very much. Yeah. I will take first the question about the omnibus, because I think the answer from my side, it's quite straightforward. I think most experts working in this field are quite concerned about the initiative of the digital omnibus.
As with the idea of simplification and deregulation, it's a first step towards dismantling some of the main safeguards that we have in our digital framework. And we are talking about, for example, some of the rules included in the GDPR, which also influence the functioning of a lot of other regulatory frameworks because, for example, the political ads regulation cannot work without the GDPR. It sets a dangerous precedent, a precedent of scaling back fundamental rights at the EU level. And it also sets, creates a dangerous precedent in terms of scaling back rules that have been just put forward because we're talking about the AI act as well. The AI act is not even fully applicable yet, and yet it's already subject to review.
To be honest, I also don't think this is ideal for, like, businesses either, even though one of the ideas behind that was to simplify the framework also for the private sector. But this creates additional legal uncertainty. So, yes, we're quite concerned, and we don't think this takes into consideration concerns around democracy and about the impact of new technologies on democracy at all. I think there was another question addressed to me on having shared benchmarks when it comes to implementation of the current regulatory framework. I'm not entirely sure what this refers to, but if it's about the implementation, for example, of the DSA and the risk assessment reports, I think we would need better guidelines addressed to very large online platforms of how to conduct these risk assessments so that they cannot just justify compliance by doing the very bare minimum and not including any useful content in the assessments.
So in that sense, I think we need clear benchmarks of what's required from the platforms in terms of obligations.
Missus Narciso.
Speaker 16
Yes. So this is a discussion, of course, as researchers when we're producing a report, we have it's a smaller level, but it's the same as EU regulation. Like, how do we define what is legitimate and what is, you know, legitimate amplifications? And, the example of the planflets was made. And it's a good example because, actually, you know, what AI allows us is to produce content massively, and nobody's contesting that.
So, like, when we invented the free press, the printing press, that changed things. The thing is that when you receive a pamphlet at the post office or from someone at your, you know, at the exit of the train station, it's another human, and you know it's another human. The problem with bots and false simplification is that these profiles are not identified as bots. And we can have a discussion which is valuable and democratic about, okay, instead of, you know, making bots something regulated and maybe illicit or illegal, we can maybe make them okay as long as they're very visible and identifiable, which was what we tried to do with AI images. But I think I'm entitled to know that if a candidate that I'm following has tons of comments saying that he is incompetent and that he's failed in his, you know, last legislation, that gives me the idea that maybe I missed something if those comments were made by real people or by fake profiles.
The same happens with the suppression. Regarding LLMs, we can have exactly the same discussion. So I would say here, we are not talking about, once again, removing content or making any editorial decision about how LLMs reply, but thinking about ways in which we could, for example, give context to any user using an LLM. So one of the things would be, can I know which LLM I'm using? So if I go to a service and they use AI, can I know if it is DeepSeek or if it is OpenAI or if it's Grok?
Because then I can understand also the editorial choices it makes for me. Can I have a banner on the right or left side of my screen saying, you know, if you want more information about vaccines, if you want more information about elections, these are the authorities that can give you information on that? So can we think about ways to give the user context on what he's seeing and make letting the user have, of course, the freedom and the liberty of deciding, but therefore getting all the information. So, I think there's a lot of things in terms of design choices that we can give users context that do not have to do with either removing content or making any editorial change.
Yes, Missus Miller.
Speaker 20
Oh, sorry. Yes. I can answer two questions. The one on sovereignty and trust in AI systems. And this is also a concern we have and that we have frequently discussed with electoral authorities who are approaching us more and more. Okay, how should we go about adopting AI systems for electoral administration?
And with which vendor should we go, and what kind of vendor due diligence should we do? And here, we see that most of the infrastructure is still in private hands. Right? So which means that private institutions have a lot of access and oversight of our personal data, which is concerning and which is something that we should do about because it's, yeah, it can impact fundamental rights, and we need to or the electoral authorities also don't have really the oversight of what happens with the data, how is it stored, how is it protected. Like they very much depend on those vendors to adequately protect those systems, which is a concern.
And so here, we say, yeah, it would be great if more of that infrastructure would be public or at least public interest and not supporting private businesses. And then, so yeah. And then on the Hungarian elections, I think there are two things here. Short term, I agree or I think we should enforce the digital rules that we have or digital legislation in the EU, meaning especially the DSA. This is something that the European Commission can do short term to make sure that the rules are followed and adhered to by anybody, including institutions.
And, but long term, I think we should really address institutional resilience. And that means it goes beyond the law. Right? It's about building long term partnerships and making sure also to have that connection with electoral authorities within the EU to make sure we're in close contact with them and to support coordination mechanisms, especially like I said, interagency collaboration is becoming more and more important, but also coordination with platforms themselves. The Philippines, for example, have shared with us that on an election night, they do a live reporting and identifying of threats by having all actors in the same room, meaning platforms, regulators, electoral authorities, civil society, journalists, so that they can respond much faster in an incident of crisis.
And this is also something that the European Center for Democratic Resilience can do, for example, and to keep the national electoral authorities close?
Javier Zarzalejos
Thank you, Mrs. Miller. Now in this second round, we will exhaust the speakers list. There will be one minute for each speaker, and then you will have the opportunity to reply. According to the list I've got here, we start with Michel de Morse, who will have one minute for your questions. Try to do your best to be as specific as possible, both in your questions and in your answers.
Christel Schaldemose
Thank you very much. I will try to be very specific because it's also, right now, we have had a lot of good questions and answers. So, I mean, we're talking about long term and short term, but what would be the best advice you could give to citizens right now facing elections, Denmark, France, Hungary, etcetera, in the light of how AI is used right now? What should you do as a citizen if you have concerns about the impact of AI in elections? Thank you very much.
Michael McNamara
Thank you very much, President. The risks posed to elections are well documented in EU observation missions, and you don't have to look outside of the European Union. We know the risks within the European Union as well. But in the light of that, the AI omnibus, as you've mentioned, proposes to delay the watermarking of AI materials without differentiating, of course, between text, images, and videos. Do you think that there is a differentiation to be made in terms of this?
In that regard, I mean, you think that, obviously, it's harder technologically to watermark or to mark text, but do you think images and videos pose more of a risk to elections? Do you think they should be prioritized in terms of watermarking, or do you think text is equally problematic? And then, what specific legal changes would you like to see? And would they be to the high risk annex, which the commission has is able to propose changes to, or would it be to the general purpose AI code of practice, which deals with democratic processes, albeit at a very, sort of high level rather than, is more granular detail required in that regard?
Javier Zarzalejos
Thank you, Mister Magnumara. Mister Piperea, one minute. Just left. He is left. Okay.
Then, mister Minshew, one minute then.
Nikola Minchev
Thank you very much, chair. So dear speakers, dear colleagues, in about a month, Bulgaria will hold its eighth parliamentary elections. For the last five years, this prolonged political instability has created fertile ground for disinformation and AI-driven manipulation. In this context, threats to democratic processes seem to form a chain of vulnerabilities. It starts from a fragile media ecosystem, moves through platform environments where manipulative campaigns take advantage of loopholes, and ends in a civic space where culture wars and disinformation overshadow democratic debate. This is why the discussion cannot remain limited to one policy field alone, and why tackling this threat effectively requires clear coordination between all our committees, Liebe Inc, COCALT, and of course, alongside the central role of AFCO and UDS.
And since these vulnerabilities are interconnected, which you already mentioned in your first intervention, that we have a very advanced framework. But still, I would like to say, beyond regulation, what concrete approaches would you recommend to strengthen our citizens' immunity to culture wars and disinformation? Thank you very much.
Javier Zarzalejos
Thank you, Mr. Mintzschev. Before I give the floor to Mr. Laszlo, let me point out that the presentation will be available only until 6. So probably the replies won't be translated.
Mr. Laszlo, floor is just one minute.
András László
Thank you very much. I think the speech of madam Tinekastriq and the reaction from one of the experts is absolutely stunning and showcases perfectly how EU interference is set up. The Russian interference article she referenced is totally unsubstantiated. No proof whatsoever in that article about Russian interference in Hungary. None.
It was published on a website funded by foreign governments, amplified by foreign media, amplified by NGOs, and it comes to the fact that a few days later, the European Commission announces that it launches its wrap it activates its rapid response system, which is basically the online censorship we have witnessed in the last couple of elections. This is how fake interference, using a fake pretext of foreign interference, is used by the European Commission to interfere in the online space right in the middle of a national election. It would be really nice also to highlight that the experts here come from NGOs who are funded by tens of millions of euros from the European Commission. No disclaimer whatsoever.
Javier Zarzalejos
You made your point. Now I give back the floor to our speakers for four minutes. Start with your replies.
Speaker 22
Thank you very much. Yes. I wanted to reply to the question about AI and watermarking and the AI Act. I will try to be fairly quick even though the topic is very technical. On watermarking, it's true that there are a lot of technical difficulties, with watermarking on the different kind of content, and we need to agree on specific solutions that would be differentiated, depending on the sort of content, whether it's video, whether it's audio, whether it's written content.
I don't necessarily think we would need to have a prioritization, but I think we shouldn't let us stop from the technical difficulties. Meaning that there are experts working on different technical solutions. So the solutions are out there. We need to adopt them. And this is also what putting forward new regulation is about, finding new solutions.
So this shouldn't stop us. About the AI act in general, whether there are elements that could be adjusted or improved. One element that I mentioned already is, it's about the high risk AI systems. So in the annex, systems intended to be used to influence elections are mentioned as well as some of the systems that would need a risk assessment. Unfortunately, it's a very restrictive notion of this sort of systems because the intentionality is very difficult to prove.
So that would benefit from a clarification so that this could be extended to AI systems that influence elections even though they were not put in place exclusively for that purpose. And I think in general about citizens and how to strengthen the immunity of citizens and so on. Something that wasn't mentioned today, but is often mentioned, is the media literacy element and digital literacy as well. I think it's very important for citizens to know how certain technologies work, what are the main risks, and also what are their main rights as well because there are rights also contained in the new regulations that they could take advantage of.
Javier Zarzalejos
Thank you, Miss Calabresse. Miss Narciso, your closing remarks or replies?
Speaker 16
Yes. Again, referring to short term solutions, not solutions, but measures that can be taken in general regarding citizens, AA, and AI. Once again, initiatives that promote or help to identify the provenance and authenticity. And this comes back to the second point, which is it is much easier to create systems that prove the authenticity of things that are human-made than it is to do the other way around. And this is going to become more and more an issue when we're entering the stage of the photocopy of the photocopy in which one day there will almost be no human made, for example, written content.
Considering that and also considering the question on tools and what we can do, there is a very big gap right now in terms of European capacity, is archiving digital evidence. So they are used by researchers looking into both Fimi and other kinds of manipulation and AI situations in elections. There is archive.org, which is owned by the Library of Congress in the US. There is an archive today, which is US based and is being actually targeted with a lawsuit. There is no European archive for digital evidence.
And I'm saying this because there may be things even happening right now that we may not be aware of and that citizens or CSOs or even law enforcement may want to archive in a way that stays for the future, in a way that preserves GDPR, for example, which these other archives do not. And it would be really important to create a system that allows to archive digital evidence, GDPR compliant, but that allows for us to look back at things that may disappear tomorrow. Thank you.
Speaker 20
Thank you. Yeah. I want to end by saying that we are sometimes very worried about or we think that AI-generated disinformation targeted or wants to pursue voters. But I think with already high quality information, it's hard to change people's mind. So maybe low-quality information does not really change that.
But what is more worrying is actually polarization, but also voter confusion. And this is also how actors can influence an election, right? If you, for example, spread wrong information about where the election takes place or how to vote, then that can already influence voters' behavior and deter voters from voting and participating in the election. So to answer the question of what we advise citizens to do is really to verify and check the official information from electoral commissions. Because they also shared with us that they find it really hard for them to get official information picked up because nowadays, many citizens inform themselves through chatbots.
And they noticed that these chatbots, AI chatbots, do not pick up their official information. So that is something I think all citizens should do too, yeah, access official websites of electoral bodies and make sure to get the right information about the election. Thank you.
Javier Zarzalejos
Thank you. Thank you, Mr. Smier. I wonder if you would like to make any closing remark, mister Gerence or Adam, you are so left. So, well, I just would like to thank all participants, both MMPs and our guest speakers for having made this event an important moment for our discussion in democracy and elections in the area in the European Union. So, I would like to thank especially my colleague, the co-chair, the vice-chair of ACCO, and the US committees as well as the chairs of AFET and INCO committees for their contribution.
And finally, I want to give our special thanks to our guest speakers, Ms. Calabrese, Ms. Narciso and Mrs. Miller for the presentation and interventions in the public hearing. Thank you very much.
Now this hearing is concluded. Thank you all.
Thank you.
Ask anything about this video — I analyse its transcript to answer.