The Workstream on Safeguarding Access to Reliable Information in the Age of AI held its third meeting on 1 October 2026. Co-chaired by Luxembourg, Spain and Ukraine, the closed-door online session brought together representatives of signatory States of the Partnership for Information and Democracy, civil society and researchers. The discussion focused on information integrity and manipulation: the technical challenges identified, the open questions for research, regulatory gaps and the policy responses emerging across regions.
Key insights from researchers, government and regulatory authorities
How AI is changing manipulation
The first half of the meeting focused on the issues at stake, looking at how AI also affects and distorts the flow of information itself through algorithmic gatekeeping, influence operations, and manipulation. It is increasingly used to sow distrust, to emphasize some views for political gain, or, as some speakers emphasized for economic gain.
- On economic gain, Aleksandra Atanasova, Reset Tech’s lead open-source intelligence (OSINT) researcher, presented Reset Tech’s latest report “Monetized Slopaganda”, emphasizing how monetisation is financing AI-driven political clickbait and deepfakes. One issue stems from the lack of transparency on what accounts earn from platform programmes and therefore makes the financial incentives behind coordinated campaigns hard to assess. According to the report, platforms have not consistently applied their stated labelling policies. Participants argued that monetisation should be established as a systemic risk factor under the Digital Services Act.
- On identification of AI driven manipulation, Dr. Vera Schmitt, Professor of Information Systems at JGU Mainz, explained that it is evolving faster than the methods used to evaluate it. The phenomenon of “LLM grooming”, described as the seeding of the web with manipulated narratives that AI systems later retrieve and reproduce, makes AI systems themselves a new target for manipulation. She stressed that AI has also lowered the cost of producing realistic deepfakes and that labelling alone is not enough, and that fact-checking may not keep pace with narratives that spread before they are flagged. She argued that existing detection tools need better coordination, integration into platforms and sustainable funding, rather than new efforts starting from scratch.
Regulatory and policy responses
The second half of the meeting focused on the current and forthcoming regulatory responses to AI-driven manipulation, with expert interventions critically analysing the European AI Act whose transparency rules for chatbots, deepfakes, and AI-generated content have taken effect since 2 August 2026, but also approaches from other legislative contexts, including Ivory Coast’s Audiovisual regulator (HACA) presenting the regional perspective of the Second Abidjan Declaration and Brazil, with its forthcoming AI bill.
On the EU AI Act, Ayşe Gizem Yaşar, Senior AI Researcher at the École normale supérieure, shared that harmful manipulation can arise not only from false information but also from the selective presentation of accurate information. She explained that the Act addresses it through Articles 5 (prohibits certain manipulative and deceptive practices) and 50 (imposes transparency obligations) but its prohibitions may be too narrowly drawn, and a focus on high-impact capabilities may overlook risks that become systemic through widespread deployment On article 50 specifically, with regard to rules on transparency and provenance Bruna Martins dos Santos, Policy and Advocacy Manager at WITNESS shared their position by stressing that marking AI-generated content should go beyond a binary “AI or not AI” label, be privacy-preserving by design and extend to open weight models
In Africa, while no specific AI legislation has been adopted, the regulators are taking a proactive approach. The second Abidjan Declaration, adopted by regulators in June 2026, extends their dialogue with platforms to AI. It calls for visible labelling of AI-generated content, fair remuneration for creators and better representation of African languages in AI systems . It provides the basis for potential regulatory frameworks. Finally Brazil presented its regulatory approach, which includes a substantial body of law relevant to AI, and a main AI Bill (No. 2.338/2023) which is still under consideration. The bill regulates AI systems rather than content moderation, and includes remuneration and opt-out rights for rights holders as well as transparency obligations
Next steps
The fourth meeting will take place on 2 December 2026. The Forum will present recommendations on harmful manipulation, drawing on the replies to its questionnaire submitted by States of the Partnership on Information and Democracy. A dedicated session testing the final recommendations will also be held at the Internet Governance Forum (IGF) in December.
