In the battle for trust, AI is a problem and can be a solution
A blog by the Frontier Tech Hub
“It’s the content that AI is producing that is increasingly difficult to differentiate from authentic content, and this is where we’re seeing the greatest risks emerging. Audiences’ confidence and ability to verify information is really decreasing.”
— FT research participant, International development charity
AI is accelerating challenges to democratic integrity
The information environment is changing fast. AI is making it cheaper and quicker to produce content at scale, and harder for people to know what to trust. Malign/hostile information interference has become a well-established pattern the world over. In a recent piece of research undertaken by the FT Hub with partners across Eastern Europe and Central Asia, we found that this is particularly problematic in countries where local-language media is thinner, regulatory governance of digital platforms is limited, and where existing societal or cultural tensions offer easy handholds for exploitation. In discussions, one participant noted:
“Mis- and disinformation is the biggest issue we deal with because it impedes progress in other areas. It’s not just a problem in itself — it works across climate, trust, cohesion. It cuts across everything.”
— FT research participant, Non-departmental organisation
Through the FT Hub’s role supporting those testing new ways of harnessing and scaling tech for good, we set out to test: how might AI be part of the movement to improve information resilience across Europe?
Specifically, can AI enable civil society organisations, culturally and contextually embedded but often under-resourced, to act faster and at a more ambitious scale to create more resilient societies, and to detect mis- and disinformation initiatives more effectively?
Building three solutions in three countries in three months
Between January and April 2026, the Frontier Tech Hub supported three civil society organisations across Europe to build and deploy AI-powered tools that aim to bolster information resilience. Each team received grant funding, hands-on innovation coaching, and access to an independent AI and ethics expert. Each started from a real, locally-identified problem.
Here’s a few examples of what they built.
-
CRI built a Lithuanian-language platform that teaches teenagers to detect AI-generated content. Using two large language models in concert, the platform generates realistic synthetic scenarios for students to critically engage with and learn to identify. Across five classroom workshops with around 110 students, 90% reported a learning benefit and 70% wanted to return for more. All teachers involved requested future access in order to integrate MAKARONAS into their lesson plans.
-
CSD built an easy-to-use graphical interface around D3LTA, an open-source tool developed by the French government’s VIGINUM agency for detecting coordinated content amplification across the web. Previously only usable by users comfortable with command-line programming, CSD’s interface opened it up to journalists, researchers, and policymakers. In April 2026, the team used it to publish an analysis of Telegram channels ahead of Bulgaria’s national elections. It detected a “dense pattern of cross-channel content circulation” as well as nodes linked to known information operations, thus demonstrating the tool’s real-world utility.
All three teams moved from concept to deployed, user-facing product within three months. All three found demand exceeded expectation, from day one. Here is what the experience taught us about the role AI can play.
Four insights into how to leverage AI for information integrity
Contextual fit is crucial for allowing technical sophistication to shine. Across all three pilots, what drove uptake wasn’t the complexity or novelty of the tech - it was how well the tool matched its context. CRI built in Lithuanian, with locally relevant examples, because teachers told them that English-language materials were useless in their classrooms. NewsMaker contextualised economic data expressed through products culturally relevant to their audience. CSD focused on making an existing analytical tool useful for existing ways of working, for the people who actually needed it. In every case, contextual fit was what users responded to.
AI can supercharge capabilities across the whole resilience spectrum. These three pilots addressed different problems at different points in the system. MAKARONAS builds audience resilience by helping young people recognise synthetic content. D3LTA UI supports detection of coordinated amplification at scale. Economix counters misleading narratives by making trustworthy information more accessible. AI proved genuinely useful in all three roles by making action faster, enabling more reach and scale, and by offering the creativity and flexibility to fit into exactly the right niches.
Responsible AI needs to be designed-in from day one. An independent review of all three pilots found a common gap: human oversight existed in practice - a journalist reviewing AI-generated content daily, facilitators present in workshops, human checks before cluster attribution - but there was a lack of intentional, upfront design and documentation of that oversight. Data governance was being discovered in motion rather than designed upfront. For these tools to be trusted, and for developing reliable and scalable tech architecture, that has to change. The lesson for future work is to embed ethics, documentation, and safeguards at the start, not after the build.
The constraints to action have shifted. It is now genuinely possible to move from idea to user-tested and deployed product in three months. That changes the question: It’s no longer whether new solutions can be built, it’s how they are sustained and how they are scaled. That is what the next phase of this work is designed to address.
Our first portfolio of pilots is part of a broader programme of work, bringing together partners across the region to understand what it takes to exponentially scale the positive impact we have proven AI can have; and to share what works - and how - with others addressing the same challenge. If you are working in this space and want to compare notes, we’d love to hear from you.
