AI Forensics

AI Forensics is a Paris-based European nonprofit that conducts independent technical audits of opaque AI-driven platforms, producing evidence of algorithmic harms that has directly triggered European Commission enforcement proceedings, Ofcom investigations, and civil society mobilization across the European Union. Founded in 2021 by Claudio Agosti, Marc Faddoul, and Salvatore Romano, it emerged from the Tracking Exposed platform-accountability project and operates as the leading civil-society technical auditing infrastructure for AI-driven platform accountability in Europe, with an annual operating budget of approximately €970,000 as of 2025–2026.

Origins: from Tracking Exposed to AI Forensics

AI Forensics grew out of Tracking Exposed (2016–2023), a project co-founded by Claudio Agosti with the Italian hacktivist community that developed open-source tools to document algorithmic curation practices on Facebook, YouTube, Amazon, TikTok, and PornHub. Marc Faddoul, who had researched recommendation systems at UC Berkeley's School of Information and worked previously at Facebook AI, joined as co-founder when the project reincorporated as AI Forensics in 2021. The reincorporation formalized the organization's focus on the emerging EU regulatory context — particularly the Digital Services Act and the AI Act — and scaled its institutional capacity to engage with European policymakers as a direct interlocutor.

Methodology

AI Forensics employs a multidisciplinary team spanning computer science, law, ethics, sociology, psychology, and communication. Because platforms typically withhold the data needed to evaluate their own algorithmic systems, the organization uses sock-puppeting, scraping, and user data donation alongside official APIs when accessible, building infrastructure that has supported dozens of independent research projects by journalists, researchers, and regulators. A €160,000 grant from the European AI & Society Fund specifically funds the use of DSA Article 40's mandated data-access framework as a civil-society auditing channel.

Key investigations

Meta political advertising. In April 2024, AI Forensics documented that less than 5% of undeclared political advertisements on Meta's European platforms were caught by Meta's own moderation systems. The European Commission cited this research when launching enforcement proceedings against Meta under the Digital Services Act — the most direct and consequential regulatory action the organization's investigations have produced.

Election integrity and AI-generated content. AI Forensics has audited multiple European election cycles: the 2024 French elections (identifying 51 instances of unlabeled AI-generated imagery, primarily from right-wing parties), the 2025 Dutch parliamentary elections (documenting coordinated manipulation and a surge in synthetic content), and the 2025 Polish presidential election (finding TikTok's labeling failures excluded 20+ million diaspora users from election information). A concurrent study of Microsoft Copilot's election-information responses found approximately one-third of answers contained factual errors.

GDPR targeting and platform safety. An investigation into X (Twitter) identified more than 30 major brands using advertising targeting based on political beliefs and sexual orientation — protected characteristics under GDPR — raising enforcement questions with European data protection authorities. A joint audit on TikTok youth mental health found the platform's algorithms routing young users to content glorifying self-harm and suicide. In May 2026, a Snapchat DSA compliance audit found the platform's legally mandated ad transparency repository unavailable 99.7% of the observation window.

Generative AI and non-consensual intimate imagery. In April 2026, AI Forensics analyzed 2.8 million messages across Telegram channels in Italy and Spain, finding nearly 25,000 users monetizing non-consensual intimate images — findings that prompted Ofcom to open a formal investigation under the Online Safety Act. A July 2026 investigation into Hugging Face found the platform hosts tools for generating non-consensual intimate imagery, with 73% of monitored requests sexual in nature. A separate analysis of Grok image generation documented weak safeguards enabling the generation of Nazi and ISIS propaganda alongside sexually explicit content.

Role in EU regulatory ecosystem

AI Forensics positions itself as the technical evidence arm of EU algorithmic accountability civil society, regularly advising EU regulators and contributing expert input to DSA and AI Act policy processes. Marc Faddoul has testified before the French Senate and the European Parliament. The organization is a registrant on the EU Transparency Register and coordinates its findings with a coalition of 80+ NGOs and experts. Its infrastructure and evidence have supported regulatory and legal work alongside organizations including AlgorithmWatch, contributing to the empirical foundation on which DSA enforcement claims are built.