{"id":3458,"date":"2026-03-27T07:49:38","date_gmt":"2026-03-27T07:49:38","guid":{"rendered":"https:\/\/scienceblog.com\/horizon\/?p=3458"},"modified":"2026-03-27T07:49:38","modified_gmt":"2026-03-27T07:49:38","slug":"an-arms-race-over-disinformation-using-ai-to-detect-ai","status":"publish","type":"post","link":"https:\/\/scienceblog.com\/horizon\/3458\/an-arms-race-over-disinformation-using-ai-to-detect-ai\/","title":{"rendered":"An arms race over disinformation: using AI to detect AI"},"content":{"rendered":"

In a world full of deepfakes and fake news, EU-funded researchers are building tools to help journalists tell what\u2019s real and what\u2019s not.<\/p>\n

By Anthony King<\/p>\n

Last winter, as Christmas markets opened across Europe, social media was flooded with alarming videos. Posts claimed that radical Islamists were \u201cinvading\u201d Christmas markets.<\/p>\n

One clip appeared to show people \u201cdisrupting\u201d the opening of the Brussels Christmas market, while a separate photo showed a market surrounded by heavy security. The message was clear: Christian traditions were supposedly under threat.<\/p>\n

But the reality was different. The videos came from peaceful demonstrations, and the photo had been generated using AI. What looked convincing at first glance was misleading \u2013 or entirely fake.<\/p>\n

This is the new information landscape.<\/p>\n

According to a recent European Commission\u00a0survey<\/a>, nearly two\u2011thirds of respondents said they had encountered disinformation or fake news within the previous week. With AI tools now able to generate highly realistic images, videos and text, it has become harder than ever to differentiate between what\u2019s real and what\u2019s not.<\/p>\n

In response, a multinational team of researchers and media specialists supported by EU funding decided to fight fire with fire.<\/p>\n

A first line of defence<\/strong><\/p>\n

In 2020, experts from universities, media houses and technology companies teamed up in a four\u2011year EU\u2011funded initiative called AI4Media. The aim was to create AI tools to help journalists and fact\u2011checkers verify digital content quickly and reliably.<\/p>\n

\u201cThere is an urgent need to develop AI techniques for the media sector,\u201d said Yiannis Kompatsiaris, research director at the Centre for Research & Technology Hellas (CERTH), who coordinated the initiative.<\/p>\n

AI has dramatically lowered the barrier to producing convincing fake content. Anyone with access to generative AI can now create fabricated images, cloned voices or realistic\u2011looking news articles. Social media platforms amplify that content at speed.<\/p>\n

\u201cWhen a fake story is supported by realistic images, it becomes much easier to believe \u2013 and more tempting to share because the content generates higher views,\u201d Kompatsiaris added.<\/p>\n

The AI4Media team built verification tools designed to fit directly into newsroom workflows. Media organisations such as Deutsche Welle in Germany and VRT in Belgium tested them in real\u2011world settings.<\/p>\n

\u201cFact\u2011checkers and journalists face suspicious images every day,\u201d said Akis Papadopoulos, a researcher at CERTH who worked on the project. He described the technology as a \u201cfirst line of defence\u201d, not a replacement for human judgement, but a way to flag potentially manipulated content quickly.<\/p>\n

\u201cIt\u2019s important to equip journalists across Europe \u2013 and globally \u2013 with tools that help them identify suspicious material fast,\u201d he said.<\/p>\n

According to the\u00a0European Digital Media Observatory, an independent, EU\u2011funded hub that monitors disinformation campaigns across all EU countries, AI\u2011generated disinformation has increased steadily in recent months.<\/p>\n

And this goes well beyond isolated hoaxes. Coordinated campaigns can influence elections, distort public debate and undermine trust in institutions.<\/p>\n

Spotting disinformation patterns<\/strong><\/p>\n

Identifying manipulated content is only part of the challenge. Understanding how disinformation spreads \u2013 who amplifies it, how narratives evolve and whether campaigns are coordinated \u2013 is just as important.<\/p>\n

\u201cWe are in a continuous loop of trying to be able to understand and catch up with the latest technology,\u201d said Riccardo Gallotti, head of the Complex Behavior Unit at Fondazione Bruno Kessler (FBK).<\/p>\n

Based in Trento, Italy, the FBK is a research centre known for its work in digital innovation, AI and the study of complex social systems. In a parallel EU\u2011funded project to AI4Media called AI4Trust, FBK partnered with universities and media organisations across Europe to analyse the wider dynamics of online disinformation.<\/p>\n

Partners included Euractiv in Belgium, Sky Italia, and the fact\u2011checking services Maldita.es in Spain, Ellenika Hoaxes in Greece and Demagog in Poland.<\/p>\n

While AI4Media focused on detecting manipulated media and integrating verification tools into newsrooms, AI4Trust built a hybrid human\u2011machine system to monitor and analyse disinformation at scale.<\/p>\n

Its platform tracks multiple social media and news sites in near real time, using advanced AI algorithms to process multilingual and multimodal content \u2013 text, audio and images.<\/p>\n

Because the volume of online material far exceeds human capacity, the system filters and flags posts that carry a high risk of being fake. Professional fact\u2011checkers then review this material, and their verified assessments feed back into the system to improve performance.<\/p>\n

The two projects are complementary. One focuses on detecting manipulated content, while the other examines how it spreads. Together, they offer both the microscope and the wide\u2011angle lens needed to understand and counter AI\u2011powered disinformation.<\/p>\n

An arms race<\/strong><\/p>\n

Using AI to detect AI might sound ironic, but it is serious business.<\/p>\n

\u201cIt is indeed funny, but it\u2019s like an arms race,\u201d Kompatsiaris said.<\/p>\n

Generative AI models are evolving at extraordinary speed. When AI4Media began, tools like ChatGPT were still in their infancy. Since then, the quality and realism of AI\u2011generated content have advanced dramatically.<\/p>\n

\u201cWe have entered a new era where the acceleration is hard for the human mind to keep up with,\u201d Papadopoulos said. \u201cTo keep up with AI, you need to be using AI.\u201d<\/p>\n

As generative models grow more powerful, detection systems must constantly adapt. That was one of the biggest challenges the researchers faced.<\/p>\n

\u201cThe technology has progressed so fast that it\u2019s difficult even for us as researchers to keep up,\u201d Papadopoulos explained. \u201cWe had to continuously update our models to detect newly generated images.\u201d<\/p>\n

The team automated parts of the verification process and regularly retrained their systems. But staying ahead demands continued investment \u2013 in both research and the media sector that depends on these technologies.<\/p>\n

The future of AI<\/strong><\/p>\n

Yet technology alone is not enough.<\/p>\n

\u201cWe need tools, but we also need policies and rules,\u201d Kompatsiaris said.<\/p>\n

Under the EU\u2019s Digital Services Act, very large online platforms must assess and mitigate systemic risks, including the spread of disinformation, and increase transparency about how their systems operate. The Artificial Intelligence Act introduces transparency obligations for certain generative AI systems, including requirements to label AI\u2011generated content.<\/p>\n

At the same time, a draft Code of Practice on transparency for AI\u2011generated content aims to encourage clearer disclosure and watermarking standards.<\/p>\n

Protecting independent journalism is another priority. The European Media Freedom Act sets out safeguards to ensure that professional media content is recognised and protected on major online platforms.<\/p>\n

Large platforms must notify recognised media outlets before removing journalistic content and explain their reasoning, giving organisations time to respond. The goal is to prevent legitimate reporting from being taken down without justification.<\/p>\n

Together, these measures and systems form a wider shield: technology to detect manipulation, regulation to improve transparency and accountability, and safeguards to protect responsible journalism.<\/p>\n

Public awareness remains just as vital.<\/p>\n

\u201cThere is no single solution,\u201d Kompatsiaris said. \u201cWe need a combination of AI tools, transparency, regulation and awareness if we want to be more effective against disinformation.\u201d<\/p>\n

This article was originally published\u202fin\u00a0<\/em>Horizon<\/em><\/a>\u00a0the EU Research and Innovation Magazine.<\/em><\/p>\n

Research in this article was funded by the EU\u2019s Horizon Programme. <\/em><\/p>\n

More info<\/p>\n