The Digital Services Act (DSA) is a set of EU-wide rules for digital intermediaries which imposes obligations on the main providers to assess the risks of the services they host and enable, and to enhance their transparency and data-sharing infrastructures. It seeks to tackle risks related to illegal content and the impact of these platforms on fundamental rights, public security, and well-being.

Entered into force in February 2023, the DSA requires Very Large Online Platforms (VLOPs) and Very Large Online Search Engines (VLOSEs) to publish a risk assessment (Art 34) and create and maintain a Public Ad Repository (Art 39).

The data sprint brought together experts from regulatory agencies, civil society, and academia to work with students in a data sprint format to shed light on how advertising works and how the data repositories are functioning, and to collaboratively develop potential policy recommendations for the successful rollout of the DSA.

Eight teams examined different risks linked to online advertisements, including: 

Health-related scam ads distributed via Meta

Building on previous work conducted by AI Forensics, this group focused on health-related scams disseminated through advertisements on Meta. Using the Meta Ad Library, the group identified potentially coordinated campaigns by focusing on near-duplicate texts, i.e., cases where multiple pages display the same ads. By conducting keyword searches, they identified more than 46,000 fraudulent health-related ads that were shown to EU users more than 292 million times.

The group's work, covered by the press and commented on by Meta, documents Meta's systemic failure to moderate its advertising ecosystems, which could have predictable negative consequences for public health and raise compliance issues under the Digital Services Act.

Read the full report here: https://aiforensics.org/uploads/meta_health.pdf

Gender Stereotypes in the Distribution of Online Employment Ads

This group studied how gender bias manifests itself in the delivery of online job Ads on Meta platforms. To explore these dynamics, the group collected 700 job ads, analyzing both their content (text, images, videos, audio) and distribution metrics. Using the INSEE classification of occupations, they examined who appeared in the ads and how audiences were targeted.

Their results revealed marked trends: men receive job advertisements more frequently than women, particularly for executive or office positions. Furthermore, the gender of the person depicted strongly influences the audience reached: ads featuring women are viewed more by women, and vice versa. Future work could include refining job and content classifications, robustness testing, and analysis of platform-specific differences to better understand the structural roots of these biases.

Political content on TikTok Ads library: The case of Romanian presidential elections

Despite TikTok's terms and conditions prohibiting political content, this group discovered how political messages ended up on the platform during the 2024 Romanian presidential election. They collected “Other Commercial Content” from TikTok's Ad library targeting Romania published in Romania between October and December 2024, downloading cover images and using optical character recognition (OCR) to extract the text embedded in the images. They searched for posts related to all presidential candidates, including Călin Georgescu, whose election was canceled due to suspected Russian manipulation.

Their findings reveal that Georgescu's TikTok campaign circumvented the platform's restrictions, promoting the candidate through content referring to elections, democracy, and national identity. The campaign's slogans and ideas circulated through these ads, even though TikTok's terms and conditions prohibit this type of content, and there is no information about who funded these videos. The group concludes that its approach could be replicated in other national contexts.

Read the full report here