Notes for the panel discussion “Artificial Intelligence: A double-edged sword for truth” @ Voices 2025

Introduction

Existing and emerging digital technologies pose significant concerns, challenges, and risks for the functioning of healthy and full democracies (see: EGE 2023). AI-driven disinformation and misinformation pose a growing threat to society’s information ecosystem. Advances in generative AI enable the mass production of fake news articles, images, and even “deepfake” videos that can be highly convincing (see: Germani et al 2024). The World Economic Forum now ranks widespread disinformation – including AI-generated false content – among the top short-term global threats (see: WEF 2024). In this challenging landscape, traditional safeguards (like regulations, manual fact-checking or content moderation) struggle to keep up with the volume and velocity of false content. A strategic response is needed that not only deploys legal or technological defenses, but also empowers people themselves to critically evaluate what they see and hear online.

AI has dramatically lowered the barrier to creating realistic false content. Deepfakes – AI-manipulated videos or audio – can fabricate speeches or events with uncanny realism, making it harder for audiences to discern truth from fiction. Propaganda campaigns can deploy swarms of AI-generated social media bots to amplify false narratives at an enormous scale. Unlike traditional misinformation, AI-driven fakes can be produced cheaply, quickly, and in bulk, potentially inundating the information space. This deluge undermines public trust in media and institutions, and can erode the shared factual basis needed for public discourse (see: Germani et al 2024; Fitz-Gerald and Padalko 2024). Against this backdrop, cultivating a critical and discerning audience becomes crucial – when technology can fake almost anything, human judgment and skepticism are a last line of defense. Strengthening critical thinking and digital literacy across society is therefore a cornerstone of mitigating the AI-driven disinformation threat.

General Overview

AI has become a double-edged sword in the political arena. On one hand, AI technologies can enhance communication and outreach; on the other, they are being weaponized to manipulate political discourse, spread disinformation, and influence elections. In Europe and beyond, AI is supercharging the scale and sophistication of misinformation campaigns. Cutting-edge generative AI can produce text, images, audio, and video that appear indistinguishable from authentic content, making it easier for malicious actors to create convincing fake political messages (see: securityconference.org).

Observers have warned that such AI-generated misinformation could become a “ticking timebomb” for democracy, with some dubbing upcoming votes as an “AI election year” (see: securityconference.org).

Several characteristics define how AI is leveraged in election interference and propaganda.

Automated social media manipulation

Bots powered by AI simulate real users to amplify political propaganda and disinformation at scale. In the past, propagandists had to program large bot networks that were often easy to spot; now AI enables more adaptive, human-like bot armies that are cheaper to run and harder to detect (see: securityconference.org).

These bots can swarm online discussions, fake grassroots support, or troll political opponents. For example, Russia has used AI to create over 1,000 fake online personas on social media, rapidly disseminating propaganda with minimal human oversight (see: csis.org; joint cybersecurity advisory).
This lowers the barrier for state or non-state actors to mount large-scale influence operations without the need for an army of human trolls.

Deepfakes and synthetic media

AI-generated deepfakes – hyper-realistic fake videos or audio – can be used to fabricate speeches or actions by politicians. Advances in AI mean that forged content can now mimic a person’s likeness or voice with alarming realism. Such content can be deployed to smear candidates, spread false statements, or create hoaxes that disrupt an election. A nightmare scenario often discussed is a deepfake video of a candidate engaging in scandalous behavior or a fake audio of a public official announcing a crisis, released at a tactically critical moment. While truly convincing deepfakes were rare until recently, the technology is rapidly improving. Even crude deepfakes can be effective if timed well – sowing doubt or confusion among voters. Notably, 90% of deepfakes online are pornographic and often target women (non-consensually placing women’s faces in explicit media) (see: securityconference.org).

This ugly fact underscores how deepfakes are also used to harass and silence female politicians and journalists, driving them out of public life as part of propaganda campaigns (see: securityconference.org).
In short, AI-manipulated media erodes the once-stable idea that “seeing (or hearing) is believing,” making citizens increasingly unsure what is real (see: misinforeview.hks.harvard.edu).

Generative text & microtargeting

AI language models can produce false news articles, social media posts, or chat responses that mimic human speech, making it easier to generate and spread disinformation at scale. Propagandists can deploy AI to write thousands of tailored messages pushing a political narrative, adjusted for different audiences or local languages. This automation can drastically amplify “fake news” websites or propaganda blogs, which may even be padded with innocuous AI-written content to appear credible while subtly seeding falsehoods (see: securityconference.org).

In tandem, AI-driven data analysis helps microtarget these messages – for instance, identifying undecided voters and delivering personalized propaganda to them. Combined with social media algorithms, such AI-curated content can create filter bubbles of misinformation, wherein users are repeatedly exposed to certain political falsehoods without counterpoints. While misinformation often reinforces existing biases rather than radically changing minds, the sheer volume and personalization that AI affords can deepen polarization and nudge the margins in tight races (see: cetas.turing.ac.uk).

Dramatic scale and speed

Overall, AI lowers the skill and cost threshold for malicious influence operations. More actors can engage in information warfare since one no longer needs advanced editing skills or fluent language ability – AI models handle those. A single operative with the right AI tools can generate a flood of fake content in minutes. For example, a researcher demonstrated that with a simple prompt, AI image generators can produce endless realistic photos of, say, “ballot boxes being stuffed” or “protesters rioting,” which could then be circulated as supposed evidence of election fraud (see: politico.eu).

In the wrong hands, this means an election can be inundated with fabricated stories and visuals almost instantaneously. Observers at the World Economic Forum ranked AI-generated disinformation among the top global risks to society, given its potential to undermine trust in institutions and factual reality (see: securityconference.org).

Erosion of trust

Beyond any single fake post or video, the cumulative effect of AI-driven disinformation is corrosive. Repeated exposure to fabricated or manipulated content contributes to a general atmosphere of distrust. Citizens begin to doubt everything they see or hear in media, including authentic news, which is sometimes called the “liar’s dividend” of deepfakes. In the long run, this undermines democratic discourse – voters can become cynical, assuming “truth” is unattainable, or dismiss real scandals as potential fakes (see: edmo.eu). Post-modern epistemologies in which objective truth is questioned or denied are an ideal breeding ground for narratives encompassing “alternative facts”.

In Europe, public concern about disinformation is very high: more than 85% of people are worried about the impact of false information on society, and 87% believe it has already harmed their country’s politics (see: edmo.eu). The rise of AI-driven propaganda only intensifies these fears.

In summary, as predicted (see: WEF 2024; Spitale et al. 2023; Germani et al. 2024) AI is being used as a force multiplier for propaganda – making disinformation more convincing, more plentiful, and faster-spreading than ever before. While recent elections have not yet seen AI “flip” results outright, the trend is clear: as the technology advances and proliferates, the threat to European democracies is growing. European security officials warn that the peak of AI-powered disruption may be yet to come (see: securityconference.org), and they urge vigilance from governments, tech platforms, and citizens alike.

Case studies in Europe

European elections in the last five years have already offered sobering examples of AI-driven interference and propaganda. Below are several notable incidents illustrating how this threat has materialized across the continent:

Slovakia 2023 – Deepfake audio sways an election?

In a parliamentary election considered a test case for AI interference, a fake audio recording emerged just two days before the vote, allegedly capturing a conversation between progressive candidate Michal Šimečka and a journalist about rigging the election. In reality, the audio was a deepfake, likely AI-generated to mimic their voices. Both targets immediately denied its authenticity, but the clip went viral during Slovakia’s pre-election “silence period” when media are barred from campaign coverage (see: misinforeview.hks.harvard.edu; edmo.eu).

The timing was pernicious: with official rebuttals muted, the disinformation spread unchallenged among thousands of Slovak voters online (see: edmo.eu). Šimečka’s pro-European party had been leading in polls, yet they narrowly lost on election day – a result many speculated was influenced by the deepfake scandal (see: misinforeview.hks.harvard.edu).
Observers dubbed this incident “the first election possibly swung by deepfakes” and “the dawn of a new era of disinformation” in Europe (see: misinforeview.hks.harvard.edu).
While other social and political factors also played a role in the outcome, the “Slovak case” stands as a stark warning: even a low-quality AI forgery, if injected at the right moment, can disrupt democratic processes and cast doubt on an election’s legitimacy.

UK & Europe 2024 – deepfake audio of political leaders

The Slovak incident is not isolated. Similar deceptive audio clips surfaced elsewhere in Europe. In Britain, an AI-generated audio of UK opposition leader Keir Starmer was circulated on social media, falsely depicting him making controversial remarks; in reality, Starmer never said those words (see: politico.eu). Fact-checkers quickly debunked the clip, but not before it had spread among partisan networks online.

These cases show how deepfake audio has become an emerging tool in political dirty tricks. Such audio forgeries are especially insidious because audio lacks the visual flaws that often give away fake images or videos; a bogus voice recording can sound entirely plausible to an average listener (see: edmo.eu).
Even when the quality is poor, as was the case in Slovakia, many people can be fooled or at least made uncertain (see: edmo.eu). Experts deem AI-generated voice clones one of the most concerning disinformation tactics currently, noting that detection is difficult and debunking often comes too late to fully counter the initial lie (see: edmo.eu).

Poland 2023 – AI voice in campaign attack ads

It’s not only shadowy anonymous actors using AI in elections – sometimes mainstream political parties have experimented with these tools. In Poland’s 2023 election, the opposition party (Civic Platform) released a campaign video that featured a deepfake voice of at the time Prime Minister Mateusz Morawiecki, generated by AI (see: brusselssignal.eu). The video spliced real footage of the PM with audio of “him” reading leaked emails, to dramatize alleged infighting in the ruling party. Crucially, the public broadcast did not initially label that the voice was AI-generated (see: brusselssignal.eu).

Once the deepfake was spotted by commentators, it sparked a backlash. Media watchdogs like Demagog (a fact-checking organization) criticized the tactic, stating “all material generated by Artificial Intelligence should be clearly marked as such”, especially during an election, calling unlabeled deepfakes “particularly dangerous for the democratic process” (see: brusselssignal.eu)
Cybersecurity experts warned that a bogus audio of a politician announcing something shocking (for example, a fake declaration of war or emergency) could trigger chaos if believed, and thus using such techniques irresponsibly is a severe risk (see: brusselssignal.eu).
Poland’s case highlights that AI propaganda is not just a foreign intrusion problem; it can be a domestic political ploy as well, raising tough questions about campaign ethics and the need for clear rules.

Far-Right propaganda images in western Europe

Across Europe, extremist groups and fringe activists have begun harnessing AI to create inflammatory fake images that serve their agendas. Since the late 2010s, far-right networks have proliferated false visuals to stoke fear on hot-button issues like immigration and crime (see: theguardian.com).

For example, in the Netherlands’ 2023 regional elections, an image was circulated of Dutch Green party leader Frans Timmermans dining luxuriously on a private jet – intended to portray him as an elitist hypocrite. In reality, the image was AI-generated (tell-tale glitches in the photo gave it away) (see: edmo.eu), but not before it spread to discredit Timmermans during the campaign.
In Ireland, an anti-immigrant agitator shared a heart-wrenching picture of a destitute white mother and her children purportedly homeless on the street – blaming migrants for taking housing. The photo was entirely fabricated by AI (none of the people were real), crafted to evoke sympathy and anger (see: edmo.eu).
In France, activists produced a fake image of masses of tractors besieging Paris, exaggerating a protest by farmers (see: edmo.eu).
In Italy, during the 2024 European elections, Salvini’s Lega Nord produced and posted fake images on Instagram, including a pregnant trans person contrasted with a traditional family, Muslims in traditional clothing burning Dante’s Divina Commedia, and people eating insects. These images were shared under the slogan “Più Italia, meno Europa” (“More Italy, less Europe”) (see: alliance4europe.eu)
Each of these AI fakes was eventually exposed by fact-checkers, but only after they had infected online discourse and inflamed tensions. Experts note that so far, AI-generated images remain a relatively small portion of political disinformation (many hoaxsters still rely on miscaptioning real photos or videos) (see: edmo.eu), but the number of AI-fakes is growing. They serve as cheap and quick content for extremist propaganda, and as the technology improves (eliminating tell-tale oddities in hands, faces, or backgrounds), these visual lies could become even more convincing (see: edmo.eu).

Russian Interference in European Elections (2019 and beyond)

Russia has long been accused of meddling in Western elections, and in recent years it has incorporated more AI-enabled tactics into its information warfare playbook. During the 2019 European Parliament elections, EU analysts found “continued and sustained disinformation activity by Russian sources” aiming to undermine the EU’s democratic legitimacy (see: politico.eu). These operations pumped out fake news on divisive issues – from immigration to sovereignty – seeking to suppress voter turnout and sway public opinion (see: politico.eu). While much of that activity relied on conventional trolling and propaganda, today Russian influence networks increasingly deploy AI tools to amplify their reach (see above, “automated social media manipulation”).

In 2022-2023, investigators uncovered a pro-Russian “bot farm” named Meliorator, that was using AI-generated profiles (complete with realistic fake photos) to pose as Americans and Europeans online and flood social media with Kremlin talking points. The campaign – disrupted by a joint operation of U.S., Canadian, and European law enforcement – showed a disturbing new capability: rapid, localized narrative laundering. With AI, Russian operators could instantly churn out messages tailored to specific countries or even regions (in local languages), and have fake “citizens” loudly promote those views (see: csis.org).
European officials noted that Moscow-backed groups also experimented with deepfake videos in some instances, though these were often low-quality and quickly debunked (see: politico.eu).
The Kremlin’s persistent efforts, combining AI innovations with traditional propaganda, demonstrate that foreign adversaries are actively testing AI-driven interference tactics on European electorates – a trend expected to continue into future election cycles.

These cases underscore a range of threats: fake audios aiming to tip elections at the last minute, AI-doctored campaign ads pushing ethical boundaries, bogus images inflaming societal divisions, and foreign powers leveraging AI for mass-produced disinformation. In each instance, there was a response – denials, fact-checks, takedowns – but often the truth struggled to keep pace with the initial falsehood. Crucially, none of these incidents definitively “decided” an election on their own. However, they did shape the information environment around those elections, fueling confusion, distrust, or false narratives. The lessons for Europe are clear: the use of AI in political influence is no longer theoretical, it’s happening now. European democracies must learn from these incidents to build resilience against even more sophisticated AI interference likely to come.

Mitigation strategies

Recognizing the growing threat, governments, tech companies, and civil society across Europe have started mobilizing to counter AI-driven election interference. A multi-pronged approach is emerging, combining policy measures, technological defenses, and public awareness initiatives. Below is an overview of key strategies in play.

Prebunking, “immunization” and critical thinking skills

Critical thinking and digital literacy: key defenses

Critical thinking and digital/media literacy are widely recognized as first lines of defense against misinformation, no matter whether “organic” or “synthetic” – i.e. AI-generated. Research shows that individuals with stronger critical-thinking skills are significantly better at identifying false or misleading information online (see: Redaelli et al. 2024). In particular, mastering concepts such as distinguishing correlation from causation, checking the credibility of sources, and understanding how data can be manipulated all help people recognize “red flags” in dubious claims (see: Redaelli et al. 2024). For example, people’s understanding and application of six key critical thinking concepts (e.g. source credibility, reproducibility of evidence) strongly correlates with their ability to spot fake social media posts (see: Redaelli et al. 2024). However, simply watching traditional instructional videos on these topics does not measurably improve participants’ ability to detect misinformation (see: Redaelli et al. 2024). This suggests that while critical thinking skills are crucial, we need more engaging and effective ways to teach and reinforce them.

Media and information literacy education is not a one-off effort but a lifelong process, as digital tactics and platforms continually evolve (see: edmo.eu). A public that is both critically and digitally literate is far more likely to analyze the information they encounter, verify sources, and make informed judgments (see: edmo.eu).

In practice, this means educational initiatives must go beyond boring rote learning – they should actively involve people in analyzing examples of misinformation and practicing responses. Innovative approaches like interactive games and “prebunking” exercises (explained below) can build habits of skepticism in a memorable way. Investing in these cognitive defenses bolsters societal resilience: citizens become less likely to be deceived by organic and synthetic falsehoods, and less likely to unwittingly spread them further.
Building a society “resilient to disinformation and capable of critical thinking” is a fundamental requirement in the AI era (see: Fitz-Gerald and Padalko 2024).

Strategies for public empowerment

Every individual has a role to play in slowing the spread of false information. Empowering the public means equipping people with the skills and tools to vet information before accepting or sharing it. Key strategies include:

Investing in media literacy education: communities and schools should provide accessible media literacy programs that teach people how to evaluate online information. This includes understanding how algorithms might feed you certain content, recognizing common disinformation tactics, and learning fact-checking techniques. Studies indicate that even basic training in digital literacy can improve people’s ability to discern true from false content (see: edmo.eu). Such education should be offered not just to students but to adults as well, as continuous learning is needed to keep pace with evolving digital media (see: edmo.eu). Libraries, community centers, and online courses can all be venues for media literacy workshops, but there is a pressing need to identify stronger and more capillary dissemination pathways.
Develop critical thinking habits: individuals should practice a mental checklist when encountering news, posts, or videos – especially if the content is provocative or suspicious. Key questions include:
- Who is the source and are they credible?
- What evidence is presented and is it verifiable?
- Could there be missing context or alternative explanations?
Healthy skepticism is important; for instance, question headlines or images that seem too sensational. Make it a habit to cross-check major claims. Cultivating these habits helps build an intuitive “radar” for misleading content. In one government public awareness campaign, citizens are advised to “Think Before You Link” – pausing to investigate the issue via other credible sources and to consider whether the content might be designed to trigger an emotional, uncritical response (see: cisa.gov). By slowing down and asking the right questions, people can break the cycle of simply reacting and sharing.
Prebunking and inoculation exercises: A proactive strategy is to expose people to small doses of misinformation tactics in order to “inoculate” them against real-world fake news. Research based on inoculation theory (see: Traberg et al. 2022; Roozenbeek et al. 2020) shows that interactive prebunking games can build mental resistance to misinformation.
- For example, the online game “Bad News” lets players step into the shoes of a fake-news creator, learning to use tricks like trolling or conspiracy theories. Playing this game for just 15 minutes significantly improved participants’ ability to recognize and resist common disinformation strategies in subsequent tests (see: Roozenbeek et al. 2020). Gamified learning of this sort works like a vaccine – by understanding how one might be deceived, the public can develop antibodies of skepticism. This approach has been effective across different countries and cultures. It represents a promising, engaging complement to traditional media literacy training.

Verify before sharing: given how easily a false story can go viral, individual responsibility in sharing is vital. Before forwarding that shocking article or retweeting a sensational claim, take a moment to verify. Look for the story on other sources – if it’s important and true, trusted outlets are likely reporting it as well (see: cisa.gov). Check if reliable fact-checking organizations (such as Snopes, PolitiFact, or AFP Fact Check) have evaluated the claim (see: cisa.gov). If the content comes via social media, be wary: ask why this content might be showing up (algorithms often push emotionally engaging posts, not necessarily accurate ones (see: cisa.gov)). Also consider the original source – is it a known news organization, an expert, a random blog, or an anonymous account? If you cannot confirm the information through credible sources, don’t share it. By exercising restraint and verification, the public can significantly reduce the spread of harmful falsehoods. Remember that misinformation often thrives on our impulsive reactions – taking a pause to double-check can make all the difference.
Recognize and resist emotional manipulation: disinformation is frequently designed to bypass critical thinking by evoking strong emotions (see: cisa.gov) – for example, a fake story may aim to enrage or shock you so that you share it immediately. Be mindful of this. If a piece of content makes you feel very angry or vindicated, treat that as a warning sign and apply extra skepticism. Scammers and propagandists know that triggering anger or fear can short-circuit our rational analysis. By staying aware of our emotional reactions, we can prevent being manipulated into spreading a false message. Take a moment to cool off and fact-check the claim before doing anything else. In short: think first, share second.
Leverage tools for verification: the public can also take advantage of the many digital tools available to help spot misinformation. Reverse image search engines can reveal if a viral photo is old or doctored. Browser plug-ins and fact-check alerts can flag known false stories. There are also emerging AI-driven tools to detect deepfake videos or to analyze text for signs of AI-generation – reliability is still unclear, especially in light of the lightning fast evolution of transformers. While these tools are not foolproof, they add an extra layer of defense. Becoming familiar with them is increasingly part of digital literacy. For example, if you see a suspicious image (say, a dramatic photo circulating after an event), doing a quick reverse image search might show that the photo actually comes from a different year or context – revealing a deception. Using such tools, combined with human judgment, can substantially reduce susceptibility to AI-fueled fakes.

It’s important to note that no single tactic is 100% effective; thus, the goal for the public is to develop an overall resilience based on the Swiss cheese model for risk management (see: thedecisionlab.com). By staying informed about how disinformation works, continually honing critical thinking, and exercising caution online, individuals become far less likely to be misled. Even as AI-generated disinformation grows more sophisticated, an empowered public with strong critical thinking skills can recognize common patterns of deceit and remain anchored in facts (see: Germani et al 2024). Building this public resilience is a long-term effort, but it is one of the most sustainable solutions to the misinformation crisis.

Strategies for media professionals

Media professionals – journalists, editors, broadcasters, and content moderators – are on the front lines of the fight against false information. Their practices can either inadvertently amplify disinformation or help contain and debunk it. The following strategies are tailored to newsrooms and media organizations to mitigate AI-driven misinformation:

Strengthen verification protocols: news organizations must adapt their verification processes to the new realities of AI-fabricated content. This means rigorously verifying user-generated photos, videos, and social posts before reporting them. Journalists should use forensic tools to check for signs of manipulation in images (such as metadata analysis or error level analysis) and employ video authentication services for suspect footage. AI can also be an ally – there are AI-powered systems being developed to detect deepfake videos or to cross-verify information across databases in real time. Media outlets should invest in these emerging technologies to aid their fact-checking. For example, The Wall Street Journal has launched an internal “media forensics” task force that brings together video editors, researchers, and technologists trained in deepfake detection (see: niemanlab.org). By deploying such multi-layered verification safeguards (again, Swiss cheese model, see: thedecisionlab.com), newsrooms can catch fraudulent content before it reaches the public.
Train and educate journalists on AI disinformation: continuous professional development is critical. Media workers need to be kept up-to-date on the latest disinformation tactics and the tools to counter them. This may involve training seminars on deepfakes and synthetic media, workshops on using OSINT (open-source intelligence) techniques to trace information, and briefings on recent disinformation campaigns. Forward-looking news organizations are already doing this – for instance, the Wall Street Journal’s team conducts newsroom-wide trainings and shares guides on detecting AI manipulations (see: niemanlab.org).
- In addition, journalism schools and programs are beginning to integrate these topics into their curricula. In one example, a graduate class at NYU called “Faking the News” teaches students how to create convincing deepfakes and fake news as a hands-on way to understand and anticipate these threats (see: niemanlab.org). By educating journalists in both the technical and editorial aspects of AI-driven fakes, media organizations ensure their staff can respond swiftly and correctly when a questionable piece of content emerges.
Collaborate with fact-checkers and technology experts: combating sophisticated misinformation is a team effort. Media outlets should build strong partnerships with independent fact-checking organizations and academic researchers specialized in misinformation (feel free to drop me a line!). This might mean integrating fact-checkers into the newsroom workflow so that they can swiftly vet claims and provide context before a story is published. It can also involve collaborating with universities or tech companies that are developing advanced verification tools. Joint initiatives can help the media stay ahead of new disinformation techniques. For example, some newsrooms work with researchers who monitor online extremist forums or foreign propaganda outlets, getting early warnings about emerging false narratives.
Practice responsible reporting: how media report on disinformation can either dampen its effects or unintentionally fan the flames. Journalists should follow best practices for reporting on false claims – for instance, avoid repeating the false claim in headlines (which can inadvertently spread it further), clearly label disinformation as false, and provide audiences with the verified truth upfront. When showcasing examples of fake or AI-generated content, media should explicitly warn that the content is false or manipulated to prevent confusion. It’s also wise to give readers context about who created the disinformation and why, when known – this transparency helps audiences understand the motive and dismiss the falsehood.
Speed is not always the priority; accuracy is. In the rush of breaking news, especially if potentially tainted by AI fakes, taking the extra moment to verify prevents larger damage to public trust. Responsible reporting extends to not amplifying fringe conspiracy theories or deepfake videos unnecessarily – coverage should be proportional and always paired with debunking and factual context.
Engage in public education and transparency: media professionals can leverage their platforms to boost public awareness about misinformation. This might include dedicated segments or articles that teach readers/viewers how to spot fake content (for example, a news program might run a short explainer on how to identify a deepfake video). Some outlets publish behind-the-scenes looks at their verification process, which not only builds trust but also educates the audience on how information is vetted. When major misinformation stories arise, news organizations can host Q&As or expert interviews to dissect how the false claim spread and why it’s false. This kind of meta-reporting turns a disinformation incident into a teachable moment for the public. Additionally, being transparent about mistakes and promptly correcting erroneous reports demonstrates accountability and reinforces media credibility – crucial in an era when people are wary of what to believe. Media that actively promote news literacy among their audience help cultivate a more discerning public, reducing the effectiveness of future disinformation attempts (see: Germani et al 2024).
Support policy and industry standards: media companies and professionals also have a voice in the broader policy arena. They can advocate for standards and tools that make identifying AI-generated content easier for everyone. News organizations might join in developing ethical guidelines for AI in media, ensuring that they themselves use AI responsibly (e.g. disclosing when AI is used in creating news content) and pushing tech platforms to take steps against malicious AI-generated fake accounts or bots. While regulation is largely the role of governments, media inputs can help shape sensible policies that protect free expression while curbing clearly harmful, fabricated content. By championing transparency and accountability measures, the media industry can help set norms that discourage the proliferation of AI disinformation.

Governments & regulators

European governments and the EU are tightening rules and beefing up safeguards to protect elections. The EU in particular has led with new regulations. In 2022, Brussels unveiled a strengthened Code of Practice on Disinformation, which for the first time explicitly targets AI-driven fakes. Major online platforms (Facebook/Meta, Google, Twitter/X, TikTok, etc.) that signed on are required to “counter deepfakes and fake accounts” or face hefty fines up to 6% of global turnover under the EU’s Digital Services Act (see: quinnemanuel.com).

This regulatory backing gives teeth to what used to be a voluntary code – if Facebook or Twitter fail to quickly remove or label AI-manipulated content, they could be found in breach of the DSA. EU officials have also pressured the political actors directly: European political parties were asked to pledge not to use deepfake content in their campaigns (see: politico.eu).
Ahead of the 2024 EU elections, The European Commission’s transparency boss, Věra Jourová asked all major European parties publicly committed to refrain from deceptive deepfakes (see: politico.eu, Twitter/X), in an effort to set a norm of “hands off” AI deception. Still, deepfakes were used (see above: “far-right propaganda images in western Europe”).
National governments are taking action too. France, for instance, passed a law in 2018 to swiftly combat “fake news” during elections – it allows judges to order the removal of false election-related claims within 48 hours and mandates transparency for online campaign ads (see: politico.eu). French regulators can force social media platforms to disclose who paid for political ads and can shut down false stories before they go viral.
Law enforcement in Europe is increasingly alert to disinformation as a hybrid threat to national security; for example, Europol’s Innovation Lab in 2022 issued detailed guidance to police agencies on detecting deepfakes and intervening before they spread in the wild (see: europol.europa.eu; cjel.law.columbia.edu).

In short, European authorities are establishing legal frameworks and emergency protocols to constrain AI abuse: from requiring labels on AI-generated political content, to improving information-sharing through the EU’s Rapid Alert System for disinformation, to even sanctioning foreign individuals who conduct influence operations. The public sector’s message is clear – election manipulation via AI is unacceptable, and there will be consequences for those who engage in it.

Policy recommendations

While progress has been made, European policymakers still face the urgent task of staying ahead of AI-enhanced threats to elections. Below are key recommendations for steps that European authorities and institutions should consider to further mitigate the risks of AI-driven political manipulation:

Mandate transparency for AI-generated political content: “sunlight” is one of the best disinfectants. Building on the EU’s efforts, policymakers should enact clear requirements that any AI-produced or AI-altered political content be prominently disclosed to the public. This could include watermarking or labeling of deepfakes and synthetic media by their creators and distributors. For example, if a campaign uses an AI-generated image or voice clip in an ad, it should carry an on-screen notice or audible disclaimer. The EU AI Act already imposes transparency obligations (developers must clearly disclose AI-generated content, see: bioid.com), but those provisions won’t fully kick in until about 2026. Therefore, interim measures are needed. The EU and national regulators should work with platforms and advertisers to enforce labeling in the short term (via codes of conduct or even emergency legislation during election periods). Furthermore, Europe could require secure provenance records to be embedded in political ads that use AI (see: cetas.turing.ac.uk) – essentially a digital signature describing when, how, and by whom a piece of content was created or edited. This authenticity-by-design approach would allow verification of genuine vs. AI-modified content. The goal is to make sure people can immediately tell when something they see is synthetic, and trace its source, before it misleads them (see: cetas.turing.ac.uk). Strict transparency rules, with penalties for those who flout them, will deter malicious deepfakes and help honest political communication thrive. An open question, deeply complex, applicable throughout all these policy recommendations is: “what is political in this context”?
Strengthen detection capabilities and rapid response: even with transparency rules, some bad actors will try to slip AI fakes into the information ecosystem. Europe needs to bolster its counter-disinformation rapid response muscle. This means investing in cutting-edge detection technologies and establishing protocols to act fast when a fake emerges. EU institutions should support research into AI that can detect deepfakes, bot networks, and generative-text propaganda in real time (feel free to drop me a line!). For instance, the EU could fund an independent AI-based fact-checking system that flags potentially manipulated images or audio as they begin trending on social media. Sharing of threat intelligence is also crucial: national cybersecurity centers, election commissions, and social media platforms should be linked in a secure alert network (an enhanced version of the current EU Rapid Alert System) to immediately circulate warnings about emerging disinformation campaigns. An Election Disinformation Incident Protocol, akin to Canada’s model (see: canada.ca; canada.ca), is a good idea to play with, where a non-partisan task force can publicly alert citizens if a foreign or significant malicious influence operation is detected during a campaign (see: Stockwell et al. 2024). Such an early-warning announcement – e.g. “intelligence agencies have identified a fake video circulating, do not be fooled” – could blunt the impact of a viral lie. Additionally, law enforcement and cybersecurity teams must practice simulated “deepfake crises” so that they are ready to technically validate or debunk suspicious media within hours and communicate findings to the press. In essence, Europe should treat major AI disinformation attempts as a form of cyber emergency, deserving swift containment efforts just like a cybersecurity incident or public health scare. The forthcoming WHO ethical guidance for infodemic management and social listening can help in understanding potential risks and setting guardrails (see: Germani et al. 2024)
Hold Platforms and propagators accountable: to change the cost-benefit calculus of running AI influence operations (such as Meliorator), there must be accountability. Policymakers should ensure that the Digital Services Act (DSA) is rigorously enforced: if platforms fail to mitigate systemic disinformation risks, the EU should not hesitate to levy tough fines. Regular audits of platform algorithms and content moderation systems (as provided for in the DSA) can assess whether Facebook, TikTok, etc., are catching AI fakes promptly. The EU could also push for algorithmic transparency – requiring platforms to explain how their recommender systems might be amplifying false content, and to adjust those systems during sensitive electoral periods. Foreign actors or domestic groups who engage in AI-driven political interference should face consequences. The EU has existing frameworks to sanction foreign cyber meddling; this could be extended explicitly to disinformation campaigns. Freezing assets or banning travel for individuals (or even state entities) linked to political interference sends a strong signal. At the national level, election laws can be updated to penalize campaigns that deliberately utilize deepfakes or bots to spread lies – for instance, nullifying a candidacy or issuing fines if a party is found responsible for egregious deepfake propaganda. Political actors should be made to understand that cheating with AI is still cheating. Alongside punitive measures, Europe should continue diplomatic efforts to establish global norms against political interference, including discouraging the malicious use of AI. While rogue states may ignore such norms, a unified public stance by democracies (through the EU, G7, etc.) increases pressure and sets expectations for responsible behavior in the digital age.
Promote public resilience and education: an educated and discerning citizenry is the best defense. European policymakers should champion expansive media and digital literacy programs focused on AI and misinformation. This includes funding research on digital literacy, on media literacy, on critical thinking; and updating school curricula to teach young people how deepfakes and bots work, and how to critically evaluate online content. For the general public, governments (in partnership with NGOs) can run awareness campaigns, especially in the lead-up to elections, to inform about the possibility of AI deception. Simple messages like “Don’t believe everything you see or hear online – it might be a deepfake” or guides on how to spot common inconsistencies can go a long way. Efforts like Google’s prebunking ads are laudable and should be amplified and supported (see: reuters.com). Nevertheless, civil society cannot and should not depend only on the “good will” of Big Tech – recent U-turns in Big Tech after the 2024 US elections clearly show how things can rapidly change. This sort of activities should be managed by democratic institutions, not by private companies. Finally, empower and fund civil society initiatives that build long-term resilience: for instance, support local journalism (which can counteract fake narratives), provide grants to innovation projects that fight disinformation, and facilitate international exchanges on best practices in media literacy. If citizens are alert and skeptical of irregular content, the impact of any AI-driven propaganda can be substantially mitigated.

European policymakers should act with urgency and unity to shore up democratic resilience against AI-enabled threats. The above measures – from transparency and detection to accountability, education, and ethical norms – form a comprehensive strategy. Europe has an opportunity to lead the world in adapting democracy for the AI age, by safeguarding political discourse and elections while upholding fundamental freedoms. The cost of inaction could be severe: a future where manipulated realities distort the will of the people. By taking proactive steps now, Europe can mitigate the risks and ensure that technology serves truth and trust, rather than deception and division.

Civil society & independent initiatives

Outside of governments, a vibrant ecosystem of journalists, fact-checkers, academic researchers, and NGOs is mobilized to defend the integrity of information. Fact-checking organizations in Europe (such as France’s AFP Fact Check, Germany’s Correctiv, Italy’s Pagella Politica, Poland’s Demagog, and many more coordinated via EDMO, the European Digital Media Observatory) are on high alert for AI-fueled hoaxes. They increasingly treat misleading AI content as a top-tier threat, rapidly debunking viral fakes and publishing explainers to inoculate the public. In the Slovak deepfake case, for instance, Slovak fact-checkers and media identified the audio as fake within hours (noting anomalies and consulting voice analysts) (see: edmo.eu), and pushed that information out to the public – though reaching everyone in time was difficult.

Fact-checkers also call out domestic political actors who cross the line, as seen when Polish fact-checkers protested the opposition’s use of a deepfake voice, demanding clear labeling and ethical standards (see: brusselssignal.eu).
Media organizations are upping their game too: newsrooms now train journalists in digital verification techniques, like how to spot signs of deepfake tampering or use reverse image search to detect AI-generated images.
Several European news agencies have even formed deepfake detection units that use specialized software to scan suspect videos for inconsistencies.
Moreover, coalition efforts are growing. Especially during elections, it was common to see collaborative fact-checking projects – e.g. cross-newsroom “war rooms” to collectively monitor and rebut viral falsehoods (this happened in France 2022, with major media pooling resources to fight election fake news (see: disinfo.eu).

Conclusion

AI-driven disinformation is a complex, evolving threat – but by centering our response on critical thinking and public empowerment, we strengthen the most fundamental defense: an informed, vigilant citizenry. The strategies outlined in this report suggest that through education, skill-building, and collaboration, we can foster a culture in which falsehoods find it harder to take root – no matter whether organic or synthetic. For the general public, this means continually sharpening our ability to question and verify information, and being mindful of what we amplify to others. For media professionals, it means adapting journalistic practices and standards to an era of AI deception, and proactively guiding the public toward truth. These human-centric approaches, paired with smart technology and supportive policies, create a Swiss cheese shield against AI-enhanced misinformation.

No solution is instant, and the effort must be sustained: incorporating media literacy in school curricula, updating training programs, investing in research, in new verification tools, and keeping the conversation about misinformation alive. Encouragingly, evidence shows that people can be taught to recognize and resist misinformation – and those skills are transferable across different topics and formats. A society that values and practices critical thinking at all levels will be much more resilient to disinformation. Indeed, national security and policy experts argue that we must build societies “resilient to disinformation and capable of critical thinking” as a bulwark against the AI-enabled information wars of our time (Fitz-Gerald and Halyna Padalko 2024).

In conclusion, by empowering individuals with critical thinking skills and holding media to high standards of integrity and verification, we can blunt the impact of disinformation. The fight against disinformation is ultimately a fight for an informed public and trustworthy information space – a prerequisite and conditio sine qua non of a healthy democracy. With the right strategies, tools, and collective commitment, we can ensure that truth has the firmest footing – even in the age of artificial intelligence.

The role of AI in election interference and political propaganda in Europe