ATLAS
← Back to News and Insights

Commission proposes election safeguards for VLOPs, including AI deepfake labels

3 min

The draft also calls for pre-election recommender testing, six-month change records and public risk reports within 30 days of each vote.

LinkedIn Post
Commission proposes election safeguards for VLOPs, including AI deepfake labels
Poland · Photo: Lāsma Artmane / Unsplash

The European Commission opened a targeted consultation on 10 November 2026 on draft revised election guidelines for very large online platforms (VLOPs) and very large online search engines (VLOSEs). The draft updates the Commission’s 2024 guidance under Article 35 of the Digital Services Act (DSA), with a focus on generative AI and increasingly sophisticated information manipulation. Poland’s Office of Electronic Communications (UKE) reported the consultation on 12 November. Feedback closes on 23 November 2026.

The underlying duty is already in force: VLOPs and VLOSEs must take appropriate and proportionate measures to mitigate systemic risks, including risks to the integrity of electoral processes, while respecting freedom of expression and the right to information. The consultation would update the Commission’s recommended approach, rather than finalise new statutory text. For the largest platforms, the Commission is primarily responsible for monitoring and enforcing the DSA’s additional obligations, while national Digital Services Coordinators supervise compliance in their Member States.

What the draft would expect

The proposed guidance sets out five areas for operational change.

- Foreign information manipulation and interference. Platforms should detect coordinated inauthentic behaviour early and dismantle the networks behind it.
- Recommender systems. Platforms should identify and assess suspicious trends before they become viral, test their recommender algorithms for manipulation before every national election, and keep records of recommender-system changes made during the six months before a vote.
- AI-generated content. Deepfakes should be clearly labelled, with the label retained when the content is shared. Illegal deepfakes of candidates should be removed quickly.
- Candidate protection. Platforms should provide dedicated channels for candidates and respond quickly to threats, harassment and impersonation, applying those measures neutrally to all candidates.
- Post-election accountability. Within 30 days of an election, platforms should report to the Commission and national authorities on how they mitigated electoral risks, and publish a public version of that report.

The draft says the measures are content-neutral. It makes no suggestion or request to delete lawful content or target a particular viewpoint. That qualification matters for compliance teams designing election controls: the proposed framework focuses on systems, coordinated manipulation, provenance and response processes, not a general obligation to remove political speech.

The Commission says the update reflects two years of experience applying the DSA and implementing the Code of Conduct on Disinformation, including its Rapid Response System. The original 2024 election guidelines already linked electoral-risk mitigation to wider DSA duties covering terms and conditions, statements of reasons, recommender systems, crisis response, transparency, independent audits, advertising repositories and data access.

Recommender-system controls

The recommender provisions are likely to be the most demanding for platform governance. The draft moves beyond a general expectation to assess systemic risk by pointing to a recurring pre-election manipulation test and a defined six-month record of algorithmic changes.

That signal arrives alongside the Commission’s separate preliminary action concerning TikTok. UKE reported on 9 November that the Commission had preliminarily found TikTok may have insufficiently assessed risks from its “For You” recommender system and its harmful “rabbit hole” effect.

In Atlas’s view, platforms operating in the EU should treat the draft as a supervisory signal to begin mapping three evidence trails: pre-election manipulation testing, change management for recommender systems and post-election reporting. The consultation is still open, and the final guidance may change, but the proposed controls point to the evidence the Commission is likely to expect when assessing whether Article 35 measures were appropriate and proportionate.

Interested parties may submit their views to the Commission by 23 November 2026.

Official source: uke.gov.pl
AIPlatform RiskAI Governance

Permanent link to this Atlas analysis

Continue with coverage connected by market, topic and operating context.

Continue with Atlas

Move from this development into the relevant research, comparison and workflow.