View source for Information Quality and Integrity – European Approaches – WS 02 2026

27 May 2026 | 09:30 - 10:30 CEST | SICCO MANSHOLT | [[image:Icons_live_20px.png | Video recording | link=https://youtu.be/2TvuJlpE9C0]] | [[image:Icon_transcript_20px.png | Transcript | link=Information Quality and Integrity – European Approaches – WS 02 2026#Transcript]]<br />
[[Consolidated_programme_2026#ws02_26|'''Consolidated programme 2026''']]<br /><br />
Proposals: #27, #41 <br /><br />
== <span class="dateline">Get involved!</span> == 
You are invited to become a member of the Session Org Team by simply subscribing to the [https://list.eurodig.org/mailman/listinfo/ws02_2026 '''mailing list''']. By doing so, you agree that your name and affiliation will be published at the relevant session wiki page. Please reply to the email send to you to confirm your subscription.

Kindly note that it may take a while until the Org Team is formed and starts working.

Check the [[{{TALKPAGENAME}} | discussion tab]] and the mailing list archive for information on the development of the session.

== Session teaser ==
The 2025 Youth DIG messages called explicitly for greater youth involvement and broad user empowerment in all things AI and algorithmic governance including the potential use of 'visible markers or labels' on AI generated content. Now, new provenance tools such as C2PA and SynthID offer to do just that, providing technological solutions for enforcing standards and ensuring compliance in line with the EU AI Act and the European Democracy Action Plan.

== Description ==

The Youth DIG messages of 2025 called for visible markers on AI content and greater youth engagement in AI transparency regulation. AI-generated content is already reshaping human cognition, blurring the lines between fact and fiction as outlined in the Rome Declaration on Media Ecology and the Cannes Declaration on the Sovereignty of Mind. We will be discussing how the proliferation of synthetic images, audio and videos online that attempt to mislead, can be combatted with the help of emerging technology with contributions from youth, the public, private and academic sectors. We will consider how diverse actors including youth are affected by these evolving algorithmic environments and where potential gaps lies in current European digital regulatory space such as the EU AI Act and the European Democracy Action Plan.

This session invites the EuroDIG community to ask whether the emerging technological solutions provide young people with the trust they are demanding as well as to consider how transparency, information integrity, and sovereignty can be best protected.

== Format == 



Order of workshop:

*Entry: As the audience arrives, a Mentimeter poll is live on screen. The audience votes as they take their seats.
*0:00–0:05 — Frances Douglas-Thomson, provides the framing of the YOUthDIG 2025 messages on AI transparency, setting out the youth mandate and the call for visible markers on AI-generated content.
*0:05–0:12 — Janice Richardson joins remotely to give an overview of the European regulatory landscape, touching on how policy is responding to young people's expectations around information integrity.
*0:12–0:19 — Pascal Schneiders joins remotely to offer the academic perspective, covering what research tells us about the effectiveness of labelling and provenance tools, and where the evidence points to gaps.
*0:19–0:26 — Camino Rojo joins remotely from Google to give the industry perspective, covering how tools like SynthID and C2PA work in practice and where the limitations lie.
*0:26–0:30 — Buffer to account for interventions running over, with Francesco Vecchi beginning to draw together the key threads before opening the floor.
*0:30–0:50 — Francesco Vecchi moderates the open question and answer session. Ayça and Gabija, present in the room, contribute during this section following the key participant interventions before open floor discussion. 
*0:50–1:00 — Smee Cujic leads collaborative message drafting with the room.

== Further reading == 
*The YouthDIG messages: https://eurodigwiki.org/wiki/YOUthDIG_2025
*Google AI technology for identifying AI generated images using SynthID:  https://blog.google/innovation-and-ai/products/ai-image-verification-gemini-app/
*The Framework Convention on Artificial Intelligence - Artificial Intelligence: https://www.coe.int/en/web/artificial-intelligence/the-framework-convention-on-artificial-intelligence
*CM/Rec(2026)1: https://search.coe.int/cm#{%22CoEIdentifier%22:[%2209125948802acd7f%22],%22sort%22:[%22CoEValidationDate%20Descending%22]}
*Steering Committee on Media and Information Society (CDMSI) and expert committees - Freedom of Expression: https://www.coe.int/en/web/freedom-expression/cdmsi-and-expert-committees
*Cannes Declaration on the Sovereignty of Mind: https://sovereigntyofmind.org/
*Rome Declaration on Media Ecology and Technology Diplomacy: https://re-imagine.eu/publication/the-rome-declaration-on-media-ecology-and-technology-diplomacy
*DSA - Code of Practice on Disinformation: https://digital-strategy.ec.europa.eu/en/policies/code-practice-disinformation

== People == 



'''Programme Committee member(s)''' 
*Frances Douglas-Thomson, MA honours student in Politics, Philosophy and Economics, at the University of Edinburgh
The [https://www.eurodig.org/get-involved/planning-process/#tab-programme-planning Programme Committee] (PC) supports the programme planning process throughout the year and works closely with the Secretariat. Members of the PC give advice on the topics, cluster the proposals and assist session organisers in their work. They also ensure that session principles are followed and monitor the complete programme to avoid repetition. 1-2 PC members have signed up to each session and  will compile the messages.

'''Focal Points''' 
*Eglė Celiešienė, Vilnius Business College (will be joining remotely)
*Gabija Skučaitė, CEO Vilnius Business College (will be joining physically)
Focal Points take over the responsibility and lead of the session organisation. They work in close cooperation with the Programme Committee and the EuroDIG Secretariat and are kindly requested to follow [https://www.eurodig.org/get-involved/planning-process/#tab-organising-a-session EuroDIG’s session principles] 

'''Organising Team (Org Team)'''

''List Org Team members here as they sign up.''
*André Melancia, Technical community, Portugal 
*Pascal Schneiders - Johannes Gutenberg University of Mainz, Germany 
*Pavlos Ferlachidis
*Francesco Vecchi, Civic AI Coordinator - Eumans, YOUthDIG 2024 
*Aldan Creo 
*Smee Cujic
The Org Team is a group of people shaping the session. Org Teams are open and every interested individual can become a member by subscribing to the mailing list.

'''Key Participants'''

*Janice Richardson, Special Advisor (Insight SA) and expert to the Council of Europe (attending remotely).
*Camino Rojo, Google (attending remotely).
*Dr Pascal Schneiders, Johannes Gutenberg University of Mainz (attending remotely).
*Frances Douglas Thomson - YouthDIG 2025 (attending in person).

*Francesco Vecchi will be the moderator for this discussion.

== Messages ==  
''Rapporteur: Smee Cujic, BSoG''

# AI and deepfakes pose a threat to information integrity. “Slopaganda” is increasingly being used during political campaigns to influence voters.
# Responsibility should not lie only with users but also with media, companies, civil society organisations, NGOs and political parties. Additionally, states, government agencies, and other relevant stakeholders carry a positive obligation.
# While labelling AI-generated content needs to serve the interest of users, it may fail to address or signal discrimination, unjust influence, social harm, or journalistic integrity. Moreover, AI content is already reshaping public beliefs and labelling it may not introduce greater clarity. It can create bias where unlabeled information is automatically perceived as high-quality or truthful, even when not wholly accurate (“implied truth effect”). At the same time, people interpret labels differently based on their background and level of critical engagement. Finally, we must raise awareness and improve human sense-making of AI-generated content, for example through media and news literacy.
# User tools for synthetic media detection are fragmented and siloed. Currently, no universal approach exists, causing these tools to fail the moment a user moves a piece of media from one platform to another.
# Cooperation between states requires a more unified approach. Because AI material is generated globally, vast differences across jurisdictions make it difficult to enforce control.

== Video record ==
https://youtu.be/2TvuJlpE9C0

== Transcript ==
'''''Disclaimer:''' This is not an official record of the session. The DiploAI system automatically generates these resources from the audiovisual recording. Resources are presented in their original format, as provided by the AI (e.g. including any spelling mistakes). The accuracy of these resources cannot be guaranteed.''

'''[https://dig.watch/event/eurodig-2026/eu-ai-act-information-quality-and-integrity-creating-resilient-citizens-ws-02-2026 The Geneva Internet Platform will provide transcript, session report and additional details shortly after the session.]'''


'''Francesco Vecchi:''' I know it’s morning after a social event, so we appreciate even more your attendance to the room. Before starting, I would strongly encourage all of you to scan the QR code that you see on the slideshow. That will redirect you to Mentimeter, where there are a few questions that would help us navigate the discussion of today. Without further ado, I am extremely pleased to moderate this session. My name is Francesco Vecchi. I am Youth League 2024, former Youth League Organizing Team 2025, and currently I’m the CVKI Coordinator at Humanist and Pan -European Movement for Popular Initiative. This year, Frances, to whom I will leave the floor in a minute, decided to continue the conversation on one of the most relevant, probably, Youth League messages of last year.

And you probably might remember from yesterday’s session, every year at EuroDig there is a youth training program called YouthDig. It ends up with some messages that are supposed to inform the political conversations at EuroDig in the very same year and possibly also in the years to come. Last year, one of these focused on calling for visible markers on AI content and greater youth engagement in AI transparency regulation. This was the main starting point for the organization of this session. As you might notice, the title is slightly different. The session is titled Information Quality and Integrity, European Approaches. Of course, it built on the messages of last year, but we also came up with some more ideas to push the discussion further.

Again, I’m extremely happy to leave the floor to Frances Douglas -Thomson She was YouthDig 2025, and this year she’s part of the EuroDig program committee. And she will give you an overview of what were the messages last year. please Frances

'''Frances Douglas-Thomson:''' perfect thanks so much Francesco yes so last year we designed some messages Francesco if you could go to the message slide perhaps so yeah I was part of Youth Dig last year and like every year Youth Dig messages covers a very very broad range of topics this one in particular I think has increased relevance today and a year post Eurodig last year and Youth Dig last year it was under the title Everything AI you can find these online they’re published alongside the Eurodig messages and so this one in particular talks about watermarking and transparency of AI synthetically generated material synthetic material online so that might be audio something visual images or videos etc so basically as a program uh committee member this year i was going through some of the proposals and i found that a lot of proposals were very relevant to this we had one from google talking about new technologies that have emerged that they are applying on their platforms that seek to basically tell users whether something is synthetic material or not and maybe what percentage of an image or a video is ai generated and this obviously is a step in the right direction to encouraging transparency in this area but of course we know that watermarking is probably not and labeling as synthetic material is probably not a catch -all sort of like a way to deal with this issue and i think that’s where this conversation can sit right so there’s obviously a tension that we also need something like increased digital literacy our information environment is rapidly changing and so the reason why i think the youth dig messages are highly relevant today is because the youth dig messages don’t just sit in a vacuum and they’re not static they should be constantly evaluated and come back to because um though we ask for things the environment changes so yeah i’m really very grateful that janice and pascal are joining us today and yeah I mean if you look at this message in particular that’s on the slide now we ask for a few things we ask that digital services and platforms use visible markers to inform when AI is operating and I want us to consider whether that’s always a good thing whether that necessarily means that we can trust material or not or whether this is able to convey harm secondly we said that ethical guidance on AI applications should be developed with input from young people third we said that AI literacy should be embedded in school curricula not just how to use AI but also how AI works how they function and what the algorithms behind it are also where human oversight must be maintained and lastly we said that AI systems should always undergo impact and risk assistance so that was basically what this message says I think today we’re going to be focusing on on the synthetic side of it and and transparency essentially at the point of consumption like what do users experience when they go to consume media and and what does that mean so I’m really excited for this conversation and I will pass back to Francisco thanks very much.

'''Francesco Vecchi:''' Thank you very much Frances um as Frances was mentioning we are going to continue the conversation with some extremely interesting speakers uh first of all let me welcome janice richardson janice is an educator researcher and author she advises governments international institutions and social media companies on digital literacy rights and citizenship she was formerly the coordinator of large -scale european internet safety and anti -bullying networks and she has extensive experience in innovative learning resource and platform development for children and adults i know that she has some slides to share so i will leave her the floor to start a conversation please janice the floor is yours janice i believe you’re still muted So, we see the slide, but I

'''Janice Richardson:''' Thank you. So, I think there is one message that I feel that you have missed. I feel that young people have a real responsibility, we all have a responsibility to also understand regulation. When we’re talking about information quality and integrity, there is one very important point. And that is, we have to be able to understand what is happening in the world. legal instrument that has come out very recently in 2025, and this is the democracy shield. The core objective is quality and integrity, stopping interference from other countries in our elections, but also building resilience of young people through digital literacy. In Europe, we have a people -centric approach, rather different than the other major regions in the world.

We believe in safety rights, values. We take a proportionate approach, which is risk -adverse, because the goal is strategic autonomy and digital sovereignty. Now, it’s very difficult to maintain such an approach when we have America on one side, the USA on the other side.

'''Francesco Vecchi:''' Excuse me, Janice.

'''Janice Richardson:''' Yes?

'''Francesco Vecchi:''' I’m asking. I’m asking you to stop for just a second, because we’re experimenting some technical hiccups in the room. So we heard the starting point of your conversation, of your intervention, but now the last few seconds were cut. maybe in the meanwhile, if you don’t mind, Janice, I will reshare the QR code for the Mentimeter so people can go on with answering those questions. Okay, Janice, while we’re covering up with the slide on the Mentimeter, we ask you to try and speak again. Janice, I have just lost you. I have just lost you. Okay, never mind. In the meanwhile, again, please, participants in the rules, can the QR code go on with answering the questions?

They will lead our conversation later on during the session of this morning. In the meanwhile, maybe I’ll ask if Janice is still hearing us or she lost us again?

'''Janice Richardson:''' No, I’m here and I can hear you.

'''Francesco Vecchi:''' Okay. Okay. Let’s go. I’ll ask Janice again.

'''Janice Richardson:''' Can you see my slides?

'''Francesco Vecchi:''' No, because I’m sharing now, but I will stop presenting my slides and then you can present yours again. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay.

'''Janice Richardson:''' Now, can you see my phone?

'''Francesco Vecchi:''' Yes, now we can.

'''Janice Richardson:''' I’m not sure it’s recorded, but one of the very big challenges with digital integrity and quality, it’s the very different approaches across three regions. We, in Europe, have a people -centric approach, very much based on safety rights values. It’s precautionary, risk adverse. The goal being strategic autonomy, digital sovereignty. However, much of the content we receive comes from the United States, where the approach is more competition, innovation -oriented, with private sector leaders leading the field and really exposing enforcement rather than imposing regulation. The goal here is to restrict market assets, to maintain the dominance of the market and to shape global standards. Whereas in China, which is the upper third declared, I’m sure you’re all on TikTok, it is territorial control over data, over platforms, over networks.

The aim being to integrate AI governance with industrial policy and national strategy. And this poses, as we know, significant implications for data quality and integrity. But as Europeans, we should also be very much aware of the legal instruments, the regulation, the directives that will help us protect our data. GDPR, we all know about that, to give us control over our personal data, to ensure transparency and rights. Digital Services Act, which aims at tech companies. to avoid manipulative harmful content. The Digital Markets Act, which aims at freedom of choice and mainly cat -sack stores providers. We should also know about the Consumer Rights Directive, what you see is what you get, the 14 -day cancellation window when we’re online shopping.

And this, of course, brings us to the AI Act, which many countries see more as a charter than any form of legal instrument. The aim is to ban dangerous uses and high risk, and the table you see there actually rates risks and what to do in case of risks. There is also the AI Framework Convention, true to all conventions, or the… The Council. of Europe, the aim is to protect human rights, democracy and rule of law. But we also have to look at the user because it’s user resilience that counts and user resilience is also an objective of the democracy shield of the European Union. And we can say there are actually seven C’s to user resilience.

First of all, control. You have to know your tools and control where you are and this relies on digital competence. But then we need a community around us and this is often a problem working with digital technology. It’s very much a single person approach, the individual approach. We have to get over this. Connectedness is extremely important. Being able to communicate and this also includes reporting, when things go wrong, which once again is an area we have to tackle. confidence, confidence in our government, confidence in these various directives and regulations that can help us and co -regulation. It’s moving so fast to keep up with it. Industry has to be involved at all levels, but with young people and all of us.

I work a lot with young people, in particular now on a project called DigiGam and a Restless Plus project, where the target is for people in prison, in vocational training and in higher education. There are many things they want to know. First of all, as was already mentioned in the messages by Frances, detecting AI -generated content. But they also want to know how AI works. What are LLMs? How can I influence them? They want more control of self -generating. They want more control of their own self -generated content, which can very rapidly change become AI -generated fake news. But there is a rising concern about the AI sycophancy and how that is challenging our well -being.

The attention economy, we need to take a closer look. What is this attention economy where getting clicks is the objective even of journalists? They want to know marketing strategies, fact -checking, because now in that summary that we get, rather than knowing the real details, and of course one thing all young people want to learn more about, all of us do, algorithm bias, discrimination. I’ll hand it back to you. I hope that was comprehensible.

'''Francesco Vecchi:''' Thank you very much, Janice. I think that it was a very good setting, the scene for the conversation of today. I’m now happy to leave the floor to, wait a second. to Pascal Schneiders. Pascal is a postdoctoral researcher at the Department of Communication at the Johannes Gutenberg University of Mainz in Germany. His research interests include news use and opinion formation in digital information ecosystems, the digital transformation of news media and media and platform governance implication. Please, Pascal, take the floor. I will also ask our remote moderator, Sue Pinnim, as main speaker, please. Thank you.

'''Pascal Schneiders:''' Yeah, thank you very much, Francesco, for introducing. And I’m happy to take part in that workshop. I would like to share some insights or short literature review of what we know about labeling content on online platforms. Mostly from communication or platform studies. And first of all, my point is about visible disclosure methods such as labels or disclaimers and their effectiveness from an audience perspective. So it’s not about invisible methods such as cryptographic watermarking because, yeah, it’s evident that people can’t see them. They are invisible and that’s more on a technical side. So when considering the design of governance measures, of course, always the question arises as how prevalent deceptive synthetic images or audios or videos are on social media platforms and what effects they have on different people.

To be honest, we know very little about that because it’s the nature of things. Those content posts are mostly hidden and deceptive, and our knowledge is mainly based on a few prominent anecdotal cases. But the pressure to act is certainly there and high because the risks of undermining people’s autonomy or people’s capabilities to distinguish different forms of public communication and capabilities to form opinions and to make well -founded decisions in the absence of interventions. And what we know, in fact, is that individuals’ video literacy is only moderately developed. that is also true for younger ones. People are quite good at assessing the trustworthiness of sources based on sources familiarity, for example. And they have also a fairly good understanding of the risks online, also related to AI.

But according to initial studies, the knowledge of AI that people have is rather limited and the ability to distinguish between information, opinion and advertorials is quite low in general. And that means there’s a greater potential for manipulation in online posts that pretend to come from established sources like journalistic ones, for example. We also know quite a little about the effectiveness of labels, how effective they are in helping people to distinguish between synthetic and non -synthetic deceptive content, and especially since there are many different ways to label content, such as source -specific labels or claim -specific ones that are attached to a particular post from a specific account. And we also know that people are quite selectively attentive to warning labels, depending on the design of the label, the usage situation, people’s attitudes, their prior experiences, their involvement in different topics, and their degree of leader literacy.

And when detected AI disclosures, in general have mostly a negative influence on people’s judgments, which indicates a negative AI -based bias. For example, people generally perceive news items labeled as AI -generated to be less trustworthy, even when the articles themselves are not rated as less accurate or biased. And knowledge of journalism is interestingly a significant one right here. So the more people know about journalism, the more critical they are of AI. And when recognizing different forms of labels people wish for and contrasting colors of size, shape, and the placement of warning icons or labels are important. And that’s one of the key factors in the identification. or recognizing labels in their context. And as we know from recent studies, users, as I already mentioned, they wish for clear, understandable, intuitive, and upfront AI disclosures.

And as we know, as far as we know from the context of journalism, users care less about the technical details of AI use so they’re less interested in which models or prompts have been used. They’re more interested in about knowing of the, yeah, or more interested in the perceived openness of journalism, especially of journalism because journalism still isn’t, yeah, central intermedia, yeah, institution that intermediates between different groups in society. So trust, and trust. generated or established by being transparent is essential here. Furthermore, people who agree with the content of a synthetic post will not be or most likely not be swayed by labels or prevented to engage with the content. That’s something that we know or that is known as confirmation bias, for example.

So, persuading people or preventing people from engaging with the content or those people who have high beliefs in that specific content which is dealt with in a certain synthetic post, that’s quite unlikely. Moreover, people often forget a piece of content that has been labeled and an increased use of interstitial warnings may cause. fatigue. So labels or using labels may lose effectiveness over time because people forget that a certain content has been labeled or over time they oversee those labels. And that means that using labels may contribute to a continued influence effect when misinformation continues to influence decision making after a correction has been issued. And there are also a number of other unintended negative consequences of using labels that we know about from misinformation research.

For example, the backfire effect when correcting something or someone only further increases the belief in the misinformation. But we also have to mention that it’s not a robust empirical phenomenon in that context. So it’s quite I’m sure of how developed or how prevalent that backfire effect might be in reality. But there’s another unintended negative consequence, the so -called implied truth effect. So when a subset of misinformation is labeled as false, people may perceive unlabeled misinformation as better than true. True. And that’s yeah, it’s not always the case in reality. So what might be held is using detailed explanations of why certain content is misleading, because that might stick in people’s minds better and making use of the so -called sandwich principle.

So embedding false information between corrections at the beginning and the end, which is true. Might be helpful, helpful to. making people aware of misinformation or deceptive AI -generated content. And additionally, making use of so -called inoculations has been found to reduce the continued influence effects and to boost cognitive resistance or resilience, as Janice has already mentioned it. And in that case, people are exposed to weakened, controlled doses of misinformation and taught about deceptive techniques using, for example, gamification methods and making use of already mentioned AI literacy measures in school. That is, instead of shielding people from every false or misleading story, people can be taught to be persuasive. Techniques used by misleading strategic actors or content such as propaganda, conspiracy, theories or disinformation.

To conclude with the recommendation, making use of such inoculation could be combined with highlighting and implementing must -be -found strategies for verified content from trusted sources, rather than trying to label all AI -generated or AI -assisted deceptive content, because that is quite impossible because of the sheer amount of content on online platforms and the unintended negative consequences that I mentioned before. And a must -be -found rule for verified, trustworthy content from trustworthy sources already exists in reality in Germany, for example, for TV platforms. So we could try to adapt that. For online platforms as well. And lastly, that raises, of course, the question of who decides which content should be flagged as positive. That certainly should be done independently of the government or the state, and it should happen at repeated intervals and according to transparent criteria and processes.

Thanks very much.

'''Francesco Vecchi:''' Thank you very much, Pascal, for your contribution to the discussion. Now, before opening the floor to an open Q &A during the morning, I just wanted to say that, as you might have found on the wiki, that one of the main points of having this conversation today is that AI -generated content is already reshaping our human cognition. It is because we are exposed to that on social networks. It is because even algorithms are just feeding us what we want to receive. And these are exactly the points that raise the concerns of the Rome Declaration on Media Ecology and the Khan Declaration on the Sovereign of Mind. Sovereign of Mind. cognitive security. I mean, there are different ways of approaching the question of information integrity because we already understand that AI generated content, and in general, AI is influencing the way we understand reality, the way we face it, and therefore the way we shape our political beliefs.

Anyway, I think it’s time to share the results of the Mentimeter. And with this, I’m happy to leave the floor again to Frances to comment on the results. And also to explain a couple of technologies that might be helpful in solving at least some of the issues related to AI generated content on the social network. So please, Frances, take the floor.

'''Frances Douglas-Thomson:''' Thanks a lot. Thank you to everyone who filled in the Mentimeter. We got a lot of responses. These are the main ones, control, security, autonomy, power, protection. I think these are interesting because they’re slightly juxtaposing, right? But yeah, you can have a look at that. I wanted to very quickly, before we turn over to the floor, we were going to be joined by Camino Rojo from Google, but she’s unable to attend, unfortunately. But the proposal that we got was talking about C2PA. and SynthID, and these are two watermarking technologies that Google has rolled out just within their platform, and I think this is interesting because one, it shows that still these technologies are slightly siloed, which speaks a bit to what Pascal was saying, that these technologies are quite difficult to actually launch at scale across platforms, and what I really have enjoyed about the points made so far by Janice and Pascal is that clearly there’s attention that exists, right?

So watermarking seeks to be a good thing, but one, is it feasible? Two, is it positive? Does it send the right message? Is it able to work alongside digital literacy to provide clear transparency signals to users? Does it empower users? Does it provide them with control and does it protect them? Or does it send the wrong message, like for example give them, as Pascal said, a misunderstanding of what is true and what is false? For example, if something’s not labeled, does that mean that it’s highly reliable? Or should we have all information critically engaged with by users online? so yes these are two technologies that Google has currently that we were going to hear about SynthID essentially is watermarking but which is harder to strip away than C2PA but it’s only for Google generated AI content so it doesn’t work for other content C2PA works by embedding provenance metadata but again if you screenshot these images re -upload compress or share via social media then this metadata disappears so currently methods of labeling are fragmented and so today I hope that we can have some contributions thinking about whether that’s necessarily a bad thing.

The second thing I want to talk about and maybe go a bit of what Janice and Pascal were saying, Janice gave us a very clear overview that regulation exists already in the EU and now it’s our role to make sure that we enforce that regulation and think about how that can realistically improve our situation as users online and people who engage with content online. So I’m going to go ahead and start with Janice. Thank you very much. Thank you very much. Thank you very much. Thank you very much. Thank you very much. I would have to agree that when you label something as content AI generated, it may unconsciously make people assume that everything that’s not is evidence and is truth, and that’s quite problematic.

And I thought the point by Janice about resilience is interesting, and I would love to hear your contributions about what you think about user resilience. Is that putting too much emphasis on users, you know, having to have control and understanding of everything online? Is that too much emphasis to put on users and young people especially in having to understand constantly and determine what is true and what is not true? So, yeah, those are questions to go out. Lastly, to do with the mentee, I really would love contributions from the floor, not necessarily questions, but also comments. One, you know, have you ever shared? Have you ever shared content online that turned out to be AI generated without knowing?

Two, what do you think the biggest threat to information integrity in Europe is? Three, do you think? AI -generated content can always be visibly labeled? And are you aware of any technologies like C2PA or SynthID? And do you think they can actually solve the problem? Additionally, who do you think should be responsible for enforcing AI labeling? Should it be governments? Should it be platforms? Should users do it themselves when they publish it? We’d love to know. Thanks so much.

'''Francesco Vecchi:''' Thank you again, Frances. And let me just build on that. Yes, we already have questions from the remote hood. So not from remote audience, but if I could make my…

'''On-site participant:''' My name is Inej. I’m here with Ustick. And I was really intrigued by the question and the comment that you made on the biggest threats for disinformation here in Europe. And I do think that AI and especially deepfakes can play a role with that. Thank you. especially because they allow for it to be easier to manipulate certain images and if I can give you a bit of context I’m from Portugal and in Portugal during the last legislative elections there was just one party who was responsible for around 80 percent of misinformation that was spread online about the elections and I’ve also seen this party for example in their own newspapers sort of that they have share images for example that were fake I don’t know if they resorted to AI but it’s possible and it would make it easier of course sorry and I think this is also one concern that we need to take into account which is the role that even political parties are also playing in this and that even themselves have a responsibility not to use these technologies to produce their content if they want to be taken seriously and if they actually care about protecting our democracies.

Thank you and sorry about the voice.

'''Francesco Vecchi:''' No worries. We know the youth big program can be intense. Thank you very much. And I would strongly encourage everyone in the room to have the same approach. So please. OK, see one hand raised there in the end.

'''On-site participant:''' Hi, I’m Hurya Mameti. I come from the Action for Democratic Society, within which it’s also a hybrid info fact checking outlet in Kosovo. We deal a lot with AI generated content, especially during the times when important things are happening like elections. And we debunk many of those content being published. One of the things that we’re seeing now is. That there are being published AI generated content. I mean, all AI, even people that do not exist. and trying to actually influence to the audience, to the citizens of Kosovo when voting. They are kind of doing these interviews, putting, publishing images. When you see, let’s say, a logo of known media but with different colors or you see different things that are being within the AI -generated content, and you see that citizens believe those, even though those interviews do not really exist and they kind of try to tend to give the information that they support a political party or a specific candidate.

But beside this, we are seeing now that it’s known like slopaganda, so it’s the weak content that is being all AI -generated published and mostly related to politicians. Like we see images or videos of politicians. Politicians dancing, something that is not normal. On the other hand… As an organization, what we’re trying to do in Kosovo is gathering all the main actors, especially from media, academia, civil society, and public institutions, to at least, since we still don’t have AI law, and, of course, we deal with AI mostly according to European Union laws. But what we’re trying to do is to at least integrate AI within internal regulations, especially to the media, but also to the media to be transparent and put a watermark whenever an image is AI all generated.

Because sometimes even if they publish an image that is AI generated just to kind of show what they want to do with the article that is being published, but at least the citizens still believe it. So another point out is that what I think that it’s more risky now is, I don’t know, The pages, especially those accounts on social media that they publish this AI -generated content, when we talk about emotional disinformation, let’s say, just to get reactions or comments from the people, but the risk is that we don’t know who stays behind them. There is no data. There are not official media registered. They just tend to represent themselves as media just by putting a .info or news and a name that people believe, and you don’t know how to actually know who stands behind them and who is the one that’s publishing all this disinformation, especially generated with AI.

'''Francesco Vecchi:''' Thank you very much for your contribution. Anyone else who would like to take the floor, please?

'''On-site participant:''' thank you very much uh my name is it the beckall i’m joining from the council of europe’s anti -discrimination department there’s a lot of food for thought and also very interesting intervention right before me i wanted to also connect to this and what Frances you were asking about uh user uh reliance and whether we’re shifting too much of this responsibility to people and i wanted to bring a point that uh perhaps it is indeed too much responsibility for individuals and and the way to go forward is to shift this responsibility more to uh state’s positive obligations to ensure a um a user safe online environment and of course not only here platforms are responsible but also when it comes to the regulation of synthetic media um private actors must act i mean we talked about google’s water marking system and there’s a lot of innovation going on but states cannot outsource this responsibility to media governance to private companies, in our opinion.

And labeling, watermarking, and moderation policies are not enough unless they’re sitting in a public interest framework. So here, we believe that the state’s role should be to define these clear rules, manage the risk assessment, and ensure the transparency, but also create independent oversight and give researchers and regulators access to information and guarantee remedies for ensuring this environment. And the Council of Europe recently published a committee of ministers’ recommendation, and this specifically addresses online safety and how online safety should be part of enabling an online environment, and it should be accessible without any discrimination, and it should be inclusive and safe and pluralistic. And here, we really enforce the idea of a positive obligation, and perhaps to connect this to an earlier point on the risks.

I think for people who are not technical experts, this conversation can get very confusing, and I myself am not a technical expert as well. But I think it’s perhaps better to view transparency as not sort of an isolated technical obligation, but also within this human rights framework. And I don’t think transparency should be treated as an end in itself, but it should serve accountability and equality and access to redress. And what I think we see with these labeling, there are some questions that I’m also curious to ask the presenters. We often present them, you know, as technical solutions to misinformation, but what happens when these systems fail? The watermarking mechanism itself can become a target for manipulation as well.

And when these failures occur, whose content should become less trans? They’re trusted. Who do we give? the responsibility to, and also whose voices become the most easier to dismiss. Because, unfortunately, labeling a content as AI -generated doesn’t necessarily remove the discriminatory impact that content creates on the online environment, and we have examples of this as well. For example, there was a research, I think, in the UK Parliament that found that AI -generated anti -immigration imagery was circulating online, but even though it was marked as AI -generated, it was nevertheless amplified by the algorithm of that platform, and it reached millions of people. So, I would also be interested, I think Pascal had mentioned that it shouldn’t be on, it should be independent from the states, and perhaps that’s a way to ensure, you know, the independency.

And besides, you know, the independent oversight of this besides any governmental or political view, but I would be curious to hear if it should only stay within the private sphere. Thank you.

'''Francesco Vecchi:''' No worries at all, I think they were extremely interesting points before opening the floor for other questions just one second, I also believe it is important to get reminded of what does AI generated mean, you know, in a context where probably every one of us is massively using AI to support their work where do you draw the line of something which is AI generated or not of course, I mean, there are some cases where it is clear that you can use the term in other cases, less clear, so how do you regulate on that and the second point is let’s not forget that within the Digital Services Act, there is the code of practice on this information namely something that every company can willingly accept to adopt within their practices, but the point is that it was published before the real age of AI, so probably part of the conversation can be also discussing how can we adapt, renew the code of practice this kind of call of practice for the context we are living in.

Before leaving the floor to the lady there, I just wanted to check if there’s any question from the room, from online. No?

'''On-site participant:''' Thank you very much. I come from Armenia, from the Prosecutor General’s office. I very much wanted to kind of continue on what the Council of Europe colleague mentioned, and also one of the questions that arose, what exactly air -generated content are we speaking about? I think it’s extremely important that the states have some role in terms of regulation, but there is a fine line between, I mean, how that interference should be regulated not to violate human rights to freedom information, the private companies, and I think it’s very difficult to understand. Thank you. balanced approach, but apparently it should be a balanced approach. But another aspect I wanted to very much raise is the cooperation amongst the states, because if the regulation differs very much from country to country, it will be extremely difficult, especially to control.

I mean, the AI -generated content may be established anywhere, and then if the regulations are very different, it’s going to be extremely difficult to have a unified approach. So perhaps cooperation amongst the states is one of the things that should be emphasized when we speak about the state’s positive obligation, because otherwise it’s going to be extremely difficult to have somewhat unified approach. So this was something I wanted to raise from that perspective. Thank you. Hello. I’m just curious. Hello. I’m Jessica from ETHDIC, and I also wanted to build on… The point that was just raised of the unified approach, I think that it’s incredibly important that we work together as European countries. And we’re very good at this already.

We work together very effectively through the DSA. We work together very effectively on the DSA to the various organizations, for example, the European Board of Digital Services Coordinator. And I think that under the AI omnibus, when all of the European regulators are empowered to enforce these AI regulations, that we can continue this great cooperation. And my further point was also about kind of the opportunities and escalating harms to do with AI generated content, that there’s great opportunities in terms of our AI innovation. There’s opportunities for economic growth, for global connectivity, for AI driven productivity. Also. For education and civic participation. There are also the harms of algorithmic amplification of toxic content of AI. gender -based image abuse and also mis – and disinformation.

And I suppose the point is that we can’t just frame these as two trade -offs, that we must frame the challenge as not choosing one over the other, but building a system of governance that maximizes both the risks, that maximizes the opportunities while preventing and mitigating the systemic risks. Thank you.

'''Co-moderator:''' So we have a question from Olga Martinez in the set, and then we have a hand from Pascal, so I will first tell the question from Olga. The question is, are we overstimulating transparency as a solution when emotionally manipulative content can still be highly effective, even if users know it was AI -generated? I guess that giving the floor to Pascal is the perfect way of getting an answer to the question, so Pascal, would you like to take the floor again?

'''Pascal Schneiders:''' Perhaps to pick up the question just from before, I think certainly a holistic, bigger picture perspective is important here because as already mentioned, AI systems influence not only individual and public decision making and opinion formation but also the production and visibility of quality of information or of public relevant information in general. So we should take a holistic governance approach to make, to ensure that news media still have a viable future and can should be financed or subsidized for example by digital also by digital platforms or other funding instruments to yeah, to ensure that we also in the future can may have access to quality information. And as we know from research, for example, by Sascha Altai and Rasmus Kreis -Nielsen and Richard Fletcher from the Reuters Institute in Oxford, ensuring and funding or strengthening news media is the best way to fight misinformation because news media or making use of news media ensures people have access to other perspectives and have access to the possibility, capability to develop their resilience and to be inoculated or to be immune against misinformation.

And research clearly shows a positive. correlation between using news media and not believing misinformation. So that means we should. strengthening news media because of that correlation and not only look at the labeling negative or misleading content, but also taking the funding perspective of journalism and news media. And in regard to the oversight governance bodies, for sure I would also agree that those states independent oversight bodies should represent the pluralism of society. So to make sure that’s not, yeah, to prevent discrimination or to prevent the acceleration of discriminant perspectives or misleading content. So that’s something that we also have or know from the public service media in Germany. They are, yeah, they have oversight bodies which more or less try to represent the heterogeneity of society.

So I think that’s a way we could also follow with regard to digital platforms and AI -generated content.

'''Francesco Vecchi:''' Thank you very much. I’m just a bit conscious about time, so we’ll get one last intervention from the very back. I think the lady raised her hand first, right? And then I think we go to the messages, so please.

'''On-site participant:''' I’m Liliana coming from North Macedonia, from the Balkan region, and NGO. So basically, sorry, I just missed the beginning of the session. I do hope that you discussed it previously as information integrity becoming a part of the cybersecurity or not. So I would want to draw the attention to the information integrity part of the cybersecurity part of the cybersecurity. to this, let’s say, interdisciplinary approach. We’re entering really a period where AI -generated, AI content is really becoming… or lowering the cost of influence operations and developing different synthetic content production. I don’t think it’s becoming already just a media issue or just a disinformation issue. It’s becoming more of a global security issue, not only regionally but globally.

So basically, I’m wondering what kind of governance models are there or tools to fight and to be around information integrity and keeping it up or protecting the information integrity as we saw it on the Mentimeter. Because in smaller states and developing digital ecosystems, we don’t have the institutional capacities that could respond to this type of threats. Was there a discussion that disinformation integrity could be part of this? Part of the cyber resilience frameworks maybe in your countries? in the region, or it could become as a different approach. I’m just wondering of the tools and the approaches that were discussed or taken. So in that segment, I would want to also emphasize what we did in order to be resilient.

We designed our own software tool where we could actually assess and measure the manipulation level of what is presented from the government itself or from the media. And in that kind of situation, we found ourselves to be in between, fighting with both or sometimes three party instead. And that is very tiring, to be honest. To a certain point, it’s tiring and it’s time consuming, and it’s not cost effective for us because it doesn’t… It doesn’t bring us projects or money, it’s just bring us the truth. which is, we say, the most valuable thing at the end, trustworthiness of information. It could bring the good case and the criminal prosecution as it should be and security at large.

However, it is pointless at the end of the day. Nobody cares about the truth anymore because whether you’re presenting it, it’s just you’re fighting with the big, but it’s a different parties and stakeholder. So what are the tools and mechanisms you are applying and what are the governance models that are already there and whether we do keep pace fast enough to grab what is going to be in the future as generative AI is really fast growing. Thank you.

'''Francesco Vecchi:''' Thank you very much as a representative of civil society i cannot but agree with most of what you said just uh very briefly let’s remind that now i mean today we mentioned the dsa of course these kind of topics are also touched by the fight against disinformation uh initiatives at the european commission so it’s like one of the tools of a foreign policy and we know that exactly cognitive sovereignty is a cause is a question also of defense and security and for again we can approach the topic from different perspectives but at the end of the day they must be intertwined if we want our messages to be relevant in any way that said i fear that we are running out of time and we still have to agree on the messages so unless maybe i don’t know if uh gabby i want to some final remarks before going to the messages or ….

'''Gabija Skučaitė:''' Hello. So my name is Gabija Skučaitė I’m from Lithuania. I’m representing Academia, a group of educational institutions. So what I’m hearing today that we live in the world, which with my colleague, Professor Mehmet Urgai, we are calling times in between where liquidity, modernity is blurred and when nothing is true or false anymore. So it is difficult to distinguish what is true. Or maybe we can label everything and still we will be not sure. So this AI -saturated environment put all humanity and society in a very different position where we need to come back to what is really humane is the ability to make sense by human capabilities of what is true and what is false.

So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. So I think that’s what I’m hearing today. very much matter in this regard, in which context I see this information, which is given to me and by whom it is given to me. So I believe that we need to put more attention to raising awareness of AI algorithms, of sense -making in the age of AI, and here academia must take a really big role in this regard because we need to educate people, young generation, to make sense of the world, which is real, actually, because we live in the real world, and the world is alive and real.

We just have this meta -environment, some virtuality, which is part of our lives. So to distinguish this is a very humane capability, which we need to embrace by ourselves, which is within us, it never went away. We just… live in a little tricky environment and I I think we will be aware of this, and the labeling will not, you know, solve everything we as humans with human capabilities will. Thank you very much.

'''Francesco Vecchi:''' So before leaving the floor to the messages, because we’re running out of time, I thank Pari Esfandiari for sharing their thoughts to the chat. I’m afraid I came a bit late, but of course Pari points out that the problem is the social trust, and especially if we live in this kind of postmodernist historical novel where no one knows where to draw the line between fact and fiction, then of course trust in institutions and trust in educational institutions first and foremost is crucial to ensure information integrity. That said, without further ado, I leave the floor to Smee for the messages. Just let me remind you that the messages of your in general and even more those of the workshops, they need to reflect broad consensus of the room.

So I will let me explain how to interact with the messages and how they were drafted, but please remind that they reflect the broad consensus of the room and please just raise your hand if you strongly object to any of the messages. Share them now. Please.

'''Smee Cujic:''' So first apologies to the person who’s way behind me since I cannot face the call room. As my colleague said, please do not pay attention to comas or the full stops. We just want to talk about what generally came out during the session. So first message, A&E face of threats for information integrity. is increasingly being used during political campaigns. I see even nodding. Thank you. Next one. Responsibility should not be only on users, but states, media, and political parties themselves. States have a positive obligation. Silence. Good. Next one, a bit longer. While labeling needs to be put in the framework of interests of… It doesn’t address discrimination, unjust interference, or social harm. AI contact is already reshaping our beliefs.

It may be that AI contact, which is labeled, does not introduce greater clarity. There is a bias where unlabeled information is automatically perceived as high quality or truthful when, in reality, it may be entirely inaccurate. At the same time, labels are interpreted differently based on personal setting and critical engagement. We also need to raise awareness and human sense making of air -generated approaches. quite a long one, but I have the content. S

'''Francesco Vecchi:''' orry, just to double check, I saw that the person over there switched on the microphone a couple of times, just wanted to say did you want to say anything?

'''On-site participant:''' Yeah, it was about the point number two would it be possible to include private companies as well in this list of people? So state media, well, technology companies?

'''Francesco Vecchi:''' Yes, I think is there anyone in the room strongly objecting again to adding private in the phrase? Okay. Okay, so a few hands raised, also people that didn’t have the chance to speak before. So in general, please, if you want to take the floor, use it through the microphone, otherwise they cannot hear you remotely, so please.

'''On-site participant:''' sorry yeah if if i’m sorry i arrived late i’m uh from european fact -checking standards network efcsn i would have lots to say but i’ve and i’ve missed that train uh just if we add in all specific actors can we mention civil society organizations in that number two because they’re unless anyone’s opposed to it i agree thank you yeah my mic mic is working um i was wondering if then in this aspect we can take out political parties i i find it a bit colloquial because i think the obligations on the political parties should anyways be determined by state obligations uh unless you’re not sure if you’re not sure if you’re not sure if you’re

'''Francesco Vecchi:''' I would disagree on this, I’m sorry but I not all political parties become really part of the state some political parties and movements are not even officially recognized and in most cases it is indeed political parties that spread misinformation so I probably would rather keep the framing as it is but if anyone else has a strong objection again we can discuss about it. OK?

'''On-site participant:''' Sorry just because we were talking about this different institutions the nation state, identity providers and the transnational organizations because somehow everybody wants to have this this this this topic together, but I’m not sure whether only the tech companies are involved in the governance topics. Even though you have objections about political parties, I guess there are three actors because all these tech companies are kind of, they are intervening our knowledge and our space so that they can have different roles. And we talked a bit about governance as a post, like labeling as post topics, but not prevention related topics and governance and oversight is not only happening with the citizens. That’s why maybe we want to bring the civil society, everything in a three -way action.

It’s like private company without mentioning private companies, but if there are any objections, then it’s okay. Can I just intervene? I think the regulations in different countries whether civil society organizations, NGOs, are considered private or not. So I think, for me, if we start counting the risk is that we may miss very many that are regulated differently in different countries. So I would suggest keeping it as general as possible but providing the functions. For example, organizations that do have regulatory functions. So this way we will ensure that we do not miss any important organization from the list. So that would be my suggestion. Governmental organizations and private, I think, and NGOs, all other types may be included in any of this.

That would be my suggestion, not to narrow down.

'''Frances Douglas-Thomson:''' So you’re saying anybody that has government power over this issue? How would you phrase it in this number two? Perfect. Should we… We’ll put that in another week and I think come back to that when we go through. Yes, if I remember correctly, there will be still time to edit the messages after the conference. So in case there are still some disagreements, just quickly the last two points just to check if there’s protocols.

'''Smee Cujic:''' Okay, so we have number four. Technological approaches are fragmented at current. Is that the technological approaches are fragmented at current? Or currently?

'''Francesco Vecchi:''' At the time being, I think. No, at the time being.

'''On-site participant:''' Is it meant that technological approaches differ from country to country? What is the idea?

'''Frances Douglas-Thomson:''' Again, across platforms, but also country to country. For example, we talked about Google, but we don’t see similar technologies emerging necessarily. Maybe we just mentioned like that.

'''Francesco Vecchi:''' Okay, while we correct that, let’s go to the fifth one. Cooperation between states requires a more unified approach. Air material is generated globally, and we see vastly different approaches across jurisdictions.

'''On-site participant:''' Can I just suggest one thing as far as the point five is concerned? Cooperation and regulatory frameworks between the countries. I think it’s very important that regulatory frameworks are also aligned to the extent possible.

'''Francesco Vecchi:''' yeah personally I love the concept of interoperability but let’s see you know not necessarily homogenous but at least interoperable with one another okay any other strong objection then I would

'''On-site participant:''' to the point four I missed the beginning the technological topics other than Google have you talked about that because there are execution time authority one of the technologies but they can be used only via the big tech companies and government can propose a policy to use these technologies but it’s very recent and I’m not sure whether anybody is aware of this technology

'''Francesco Vecchi:''' Okay, then I thank you all for participating I think that the messages are extremely interesting and the conversation itself was definitely rich Thank you again for coming here today and let’s keep in touch also if you want to propose little changes in the next few days there will be still time to correct the commas, but I’m also happy to get to a broad consensus on the topic. Thank you again.



[[Category:2026]][[Category:Sessions 2026]][[Category:Sessions]][[Category:Workshop 2026]]