The European Union is urging online platforms to do more to identify AI-generated content as elections approach. The concern is that realistic images and other material made with generative AI could help spread disinformation and undermine public debate.
Vera Jourova, the bloc’s values and transparency commissioner, said existing platform efforts were a start, but called for stronger safeguards. The EU is leaning on its voluntary Code of Practice on Disinformation while broader rules are still being negotiated and are not expected to take effect for a couple of years.
Why the EU is pressing for action now
Jourova welcomed steps by some platforms to tell users when online content has a “synthetic origin.” She also warned that realistic AI products have a high potential to create and circulate disinformation, with particular risks during elections.
She urged platforms to stay alert and provide effective election safeguards. The commissioner also said she was meeting representatives of OpenAI, the maker of ChatGPT. OpenAI was not a signatory to the EU’s Code of Practice at the time described in the article.
The code is voluntary, but the Commission wants signatories to act proactively on disclosures about deepfakes and other AI-generated content. The expected EU AI Act is still in draft and under negotiation. The source says it is expected to require user disclosures from makers of generative AI technologies, but not to apply for a couple of years after adoption.
The code also connects to the Digital Services Act. The Commission has said that following the code can be a favorable signal for larger platforms subject to the DSA, which requires them to assess and mitigate societal risks associated with their algorithms, including disinformation.
What major platforms reported
Reports covering January to June were published by signatories to the code. The EU described this batch as the most extensive reporting produced since the code was set up in 2018. The code had 44 signatories, including platforms, search services, advertising industry entities and civil society organizations involved in fact-checking.
Google said it was working on watermarking, metadata and other techniques for its generative models. It also planned to use the IPTC Photo Metadata Standard to tag images generated by Google AI, so that a label could appear in Search. The report said creators and publishers would be able to add similar markup to their own images.
Microsoft described safeguards for Bing that included pre-launch testing, a phased release, classifiers, defensive search interventions and enhanced reporting. Its report did not commit Bing Search to specific additional steps over the following six months, saying it would continue reviewing its policies and practices.
TikTok said users must disclose when realistic scenes in their content are AI-generated or manipulated. It also planned to strengthen enforcement of its synthetic media policy and explore tools to improve detection, enforcement and user education.
Meta said it wanted to work with government, industry, civil society and academic partners on solutions to AI-generated misinformation. It also described plans for a Community Forum on Generative AI and participation in a working group on AI-generated disinformation under the EU code.
Transparency is part of a wider response
Labels can help users recognize that content was generated or changed using AI. The platforms’ reports point to other parts of the response too: detection, enforcement, reporting processes, media literacy and cooperation with outside organizations. These steps address different parts of the problem, from identifying suspicious material to helping people assess what they encounter.
The reports also show that commitments vary. Some platforms described specific measures already in place or planned, while Microsoft’s report emphasized ongoing review without listing new Bing steps for the next six months. Jourova’s message was that signatories should keep building on their efforts, especially as national and EU elections approach.
Elections and coordinated influence
Jourova singled out Kremlin propaganda and the risk of Russian election interference. She said platforms should account for the war against Ukraine and the upcoming EU elections, warning that malicious actors could try to exploit platform design to manipulate public debate.
The article also cites the Commission’s early analysis of the reports: YouTube shut down more than 400 channels between January and April 2023 that were involved in coordinated influence operations linked to the Russian-state sponsored Internet Research Agency (IRA).
For the EU, the immediate challenge is to get platforms to apply safeguards before binding rules take effect. The voluntary code is intended to encourage action in the interim, while the DSA already places risk-mitigation obligations on very large platforms and search engines. How platforms respond around elections will test whether their disclosures, policies and enforcement can keep pace with the risks the Commission has identified.