The Verification Crisis: Expert Perceptions of GenAI Disinformation and the Case for Reproducible Provenance
Organizations: Frankfurt University of Applied Sciences Frankfurt am Main, Germany · IMT Atlantique, UMR IRISA, Chaire Cyber CNI Rennes, France
Abstract
The growth of Generative Artificial Intelligence (GenAI) has shifted disinformation production from manual fabrication to automated, large-scale manipulation. This article presents findings from the first wave of a longitudinal expert perception survey (N=21) involving AI researchers, policymakers, and disinformation specialists. It examines the perceived severity of multimodal threats -- text, image, audio, and video -- and evaluates current mitigation strategies. Results indicate that while deepfake video presents immediate "shock" value, large-scale text generation poses a systemic risk of "epistemic fragmentation" and "synthetic consensus," particularly in the political domain. The survey reveals skepticism about technical detection tools, with experts favoring provenance standards and regulatory frameworks despite implementation barriers. GenAI disinformation research requires reproducible methods. The current challenge is measurement: without standardized benchmarks and reproducibility checklists, tracking or countering synthetic media remains difficult. We propose treating information integrity as an infrastructure with rigor in data provenance and methodological reproducibility.
Figures & tables
| Role Category | Count | Representative Titles |
|---|---|---|
| AI Researchers / Engineers | 9 (42.9%) | “VP @ social media company”, “Professor Cybersecurity”, “AI researcher” |
| Policymakers / Regulators | 2 (9.5%) | “Member of the European Commission”, “Architect of DPI frameworks” |
| Journalists / Civil Society | 3 (14.3%) | “Spiegel-Kolumnist”, “Disinformation/fact-checking professional” |
| Industry Executives | 3 (14.3%) | “Exec of a leading AI company”, “Partner at an AI fund” |
| Other | 4 (19.0%) | “Teacher”, “IT Architect”, “Ethicist/Legal Scholar” |
| Modality | Mean | Key Expert Insight |
|---|---|---|
| Deepfake Video | 6.4 | “The ultimate tool for eroding the ‘shared objective reality’ required for democracy…It creates an environment of total doubt.” |
| AI-Generated Text | 6.1 | “Large-scale text generation is the most significant threat to eroding general veracity discernment…The marginal cost…is trending toward zero.” |
| Voice Cloning (Audio) | 5.9 | “Audio cloning presents an immediate, hard-to-detect vector for fraud…Voice-based payment scams are the new phishing.” |
| AI-Generated Images | 5.8 | “It helps calibrate public expectations. People need to understand that they can be fooled.” |
| Strategy | Mean | Primary Limitation Cited |
|---|---|---|
| Media / Public Literacy | 4.8 | “Reaches only tech-savvy audiences”; “Risks increasing general distrust”; “Impact fades once novelty wears off.” |
| Digital Watermarking / C2PA | 4.2 | “Easily bypassed by adversarial content” (e.g., stripping metadata); “Coordination…massive challenge.” |
| Government Regulation | 4.1 | “Pacing problem” (Laws are too slow); “Stifles open-source innovation”; “Europe is regulating itself out of relevance.” |
| Platform Enforcement | 3.5 | Business model conflict (“Platforms profit from engagement, not truth”). |
| Technical Detection Tools | 3.4 | “The ‘cat and mouse’ dynamic…detection becomes statistically harder”; “Unreliable.” |
Generative AI Evaluation
Generative AI evaluation measures the quality, reliability, diversity, and real-world usefulness of generated outputs and AI-assisted workflows.
See all 313 papers