Researchers use ad creatives to infer tracker-advertiser links
Researchers Maaz Bin Musa and Rishab Nithyanand have developed ATOM, a novel, generalizable technique to infer data sharing relationships between online trackers and advertisers. Unlike prior methods, ATOM is independent of ad delivery protocols or specific artifacts, instead leveraging personalized ad creatives to detect when blocking a tracker affects an advertiser's ability to deliver targeted ads. This research aims to provide a tool for auditing compliance with privacy regulations like CCPA and GDPR, which require disclosure of data sharing partners.
Key Takeaways
- ATOM uses personalized ad creatives, not bid values or protocol-specific artifacts, to infer tracker-advertiser data sharing.
- The test deployment analyzed nine advertisers and reported 100% model accuracy for OpenX, Exponential Advertising Intelligence, and The Trade Desk.
- Table 3 linked OpenX to Oracle and Alphabet, Pubmatic to Alphabet, and Media Math to OpenX and Facebook.
- The researchers generated 5.3M ads, with 31.5K unique creatives, across 5,400 personas and six interest groups.
- ATOM validated some inferences with CCPA disclosures, KASHF bid analysis, and cookie-syncing logs.
Why It Matters
ATOM gives auditors a way to test whether advertisers and trackers are sharing data even when the exchange happens server-side or through other paths invisible to a browser. That matters because the paper ties the method directly to disclosure rules in GDPR, CCPA, and CPRA, and shows it can surface relationships missed by earlier header-bidding or retargeting-based approaches. The broadest signal to watch is whether auditors can reproduce ATOM’s advertiser-level results, especially the nine models that cleared the paper’s 60% holdout threshold.
Read full article at petsymposium.org