AI Forensics

Snapchat's DSA Ad Transparency

Snapchat's ad transparency tools fell far short of DSA requirements: the Ad Gallery was searchable only by advertiser name, the repository was unavailable 99.7% of the time observed, and commercial communications were not labelled on the web.

Harassment as Infrastructure: How Telegram's design enables TFGBV

Analysis of 2.8M messages across 16 groups/channels found organized networks of ~25,000 users across Italy and Spain spreading and monetizing non-consensual intimate images (NCII), including CSAM.

Artificial Elections 2.0: Generative AI in the 2026 French Elections

AI-generated content in the municipal elections remained largely unlabelled and increasingly normalized, expanding to cartoonish imagery and most prominent on the right/far right.

Grok Unleashed: An Update

Follow-up found a drop in minimally-attired images by mid-January (suggesting new safeguards), but Grok.com retained the capability and synthetic media depicting minors was still found.

Dutch Parliamentary Elections 2025 Report

Election-integrity monitoring found coordinated manipulation, new disinformation tactics, a surge in AI-generated content, and clear failures in platform moderation.

Grok Unleashed: Flood of Sexualized Images of Women and Minors

Analysis of 20k+ Grok images and 50k prompts found weak guardrails: 53% minimal attire, 81% women-presenting, ~2% appearing to be minors, 6% public figures, plus Nazi/ISIS propaganda.

Prompt, Upload, Repeat: Agentic AI Accounts Flood TikTok

Uncovered 354 'Agentic AI Accounts' responsible for the majority of AI slop on TikTok, together amassing 4.5B+ views and 43,000+ near-entirely AI-generated posts of misleading/harmful content.

Truth be Dammed: Climate Disinformation on YouTube and TikTok

With Maldita.es: dis/misinformation videos about Spain's Valencia (DANA) floods were amplified on both platforms and remained online a year later; YouTube lacks a dedicated policy and TikTok fails to enforce its own.

From FYP to WW3

During the 2025 NATO Summit, the For You Pages of 12 Dutch accounts prioritized military content (40%) and war speculation (19%); both feed and search reflected a largely pro-NATO stance.

YouTube's Safety Features Lost in Translation

Across all 83 supported languages, safety panels were unevenly distributed: English had full coverage while many languages, including European ones, lacked basic safety features.

AI Generated Algorithmic Virality (AI Slop)

Across three European countries, 25% of TikTok's top search results contained synthetic AI imagery (80%+ from 'Agentic AI Accounts'); only about half of AI content was properly labeled despite DSA commitments.

App Stores & Meta's Porn: An Uncomfortable Affair

In Spring 2025 Meta approved 12,034+ pornographic ads, shown 12.8M times across all 27 EU countries, apparently violating store and Meta policies; the same images were auto-moderated when posted as normal Instagram posts.

When Personal Becomes Profitable: Sensitive Targeting on X

X let advertisers target users by sensitive personal data (political beliefs, sexual orientation); 30+ major brands were found doing so, raising GDPR and DSA concerns as VLOPs are barred from such targeting.

TikTok's Research API: Problems without Explanations

Missing metadata for official TikTok videos, ads, and prominent accounts makes the research API unreliable for data-donation studies and algorithmic auditing.

TikTok's Polish Elections Labels: Only Sometimes, And Only For Some

Systematic failures applying election-information labels during Poland's 2025 presidential election; 20M+ Polish diaspora users were excluded from labels while fraud allegations and AI disinformation spread unlabeled.

Meta's Failing Ad Moderation: Health Scams Targeting EU Users

Identified 46,000+ ads promoting unapproved drugs and deceptive health claims, reaching EU users 292M+ times, violating Meta's own standards and potentially several DSA provisions.

Pay-to-Play: Meta's Community (double) Standards on Pornographic Ads

3,000+ pornographic ads (including AI-generated media and celebrity deepfakes) were approved and generated 8M+ EU impressions, content that would be removed if posted by regular users.

Artificial Elections: Generative AI in 2024 French Elections

Found 51 instances of synthetic imagery, primarily used by right-wing parties to dramatize sensationalist narratives during the 2024 elections; none were labeled as AI-generated.

TikTok's 'Others Searched For' Feature

The search-suggestion feature showed clickbait-style suggestions and redirected users to competing political parties (particularly AfD), potentially distorting political perception among young users.

Light on Safety: How TikTok Lite Sacrifices User Protections

With Mozilla: TikTok Lite (widely used in Global Majority countries) lacks content filtering, screen-management tools and harmful-content warnings, leaving 1B+ users exposed despite these being technically feasible.

No Embargo In Sight: Meta Lets Pro-Russia Propaganda Ads Flood the EU

A moderation loophole let pro-Russian propaganda run on Meta; under 5% of undeclared political ads were caught, and one campaign reached 38M+ users in France and Germany. The European Commission opened proceedings citing this report.

TikTok's Role in Youth Mental Health

With Amnesty International: a two-part audit found TikTok's recommendation system could rapidly lead young users toward content glorifying self-harm and suicide.