Disinformation & Trust
AI fakes and recommendation algorithms make it hard to tell what is true; trust in information is eroding.
Open in the interactive tree →Deepfakes (image, voice, video) can be produced cheaply in bulk, and platforms earn from attention. Studies find that false news spreads faster than true news (Vosoughi et al., Science 2018). Regulation such as the DSA and the AI Act's labeling duties addresses symptoms, not the incentives.
As of October 2026
The Reuters Institute's Digital News Report 2026 (16 June 2026) finds 62% of respondents worried about what is real and what is fake online, up 4 points, and overall trust in news at 37%. Researcher data access under the DSA applies since 29 October 2025 and the AI Act's deepfake labelling duty since 2 August 2026, but provenance standards such as C2PA are not yet widely deployed.
What is missing
- Tamper-proof provenance for images, video and text (e.g. C2PA, watermarks) in all devices and platforms
- Access for independent researchers to platform data to measure effects
- Reliable detection of AI-generated content (today error-prone)
- Media and source literacy across the whole population
- Business models that do not live off outrage and attention
Becomes possible once solved
- Court-proof digital recordings
- Trustworthy news and elections
- Less fraud with faked voices and videos
Open steps
- Provenance on all devices and platforms Low AI leverageGet tamper-resistant content provenance (C2PA credentials, watermarks) into cameras, editing tools and platforms, and make the labels understandable.
- Detecting AI-generated media Medium AI leverageBuild detectors that keep working on new generators and compressed real-world media, or find robust alternatives such as watermark verification.
- Scaling fact-checks and context notes High AI leverageTest whether AI-written, human-rated context notes and fact-check dialogues reduce belief in false claims at scale and across political groups.
- Measuring effects with platform data Medium AI leverageUse DSA researcher data access to measure how recommendation systems and AI fakes change beliefs and trust, with audited, reproducible methods.
- Ranking that rewards accuracy Medium AI leverageTest ranking and business models that reward accuracy or cross-group agreement instead of outrage, and measure the effect on trust.
Where AI could help
Medium AI leverage. AI can scale fact-checks and ranking research, though detection stays unreliable; business models and media literacy remain key.
- Personalized, evidence-based fact-check dialogues that measurably reduce false beliefs
- Draft and rank context notes on misleading posts, with humans rating which are shown
- Check provenance and watermark metadata at scale (C2PA)
- Flag coordinated inauthentic campaigns across platforms once researchers get data access
Shown so far
- In September 2024 a Science study with over 2,000 conspiracy believers found that dialogues with GPT-4 Turbo cut belief by about 20% on average, an effect still present after two months. source
Prerequisites
- Source-Critical History (Ranke)1824Source criticism is the classic method for judging how reliable information is
- Behavioral Economics1979Cognitive biases explain why false, emotive news spreads on attention platforms
- Platform Economy1995-2009
- Social Media & Algorithmic Feeds2004-2012Ranked feeds and viral sharing carry false claims and strain shared trust
- Large Language Models2022
- Platform Regulation (DMA/DSA)2023-2026
- Digital Identity & Trustopen