Thousands of Deceptive Ads Evade Google's Safeguards, Study Finds

Thousands of deceptive advertisements have slipped through Google's advertising safeguards, exposing internet users to fake virus warnings, misleading software promises and potentially dangerous downloads, according to new research from NYU Tandon and Radboud University in the Netherlands.

Using an AI tool they developed called AdLens, the researchers examined nearly 189,000 versions of software-related advertisements from Google's publicly accessible Ads Transparency Center and identified thousands of deceptive ads that collectively generated more than 100 million impressions across European markets.  

Some of the Google ads it identified remained active for years, while others continued running even after being reported to the company.

In one case, Google removed an advertisement linked to a known malware operation but left 41 others directing users to the same website. The findings raise questions about how effectively Google detects harmful advertising and whether its enforcement systems address entire campaigns rather than individual ads.

“Google has systems in place to screen advertisements, but we found thousands of deceptive ads that still reached users, sometimes for years,” said Ritik Roongta, NYU Tandon Computer Science and Engineering (CSE) PhD student and lead author of the study. “Even when we reported ads linked to known malware campaigns, removing one didn't necessarily stop dozens of others promoting the same threat. We shouldn't have to rely entirely on advertising platforms to tell us whether their safeguards are working.”

The researcher team’s study, AdLens: Efficient Detection of Deceptive Software Ads, will be presented on November 18 at the 2026 ACM Conference on Computer and Communications Security (CCS) in The Hague, Netherlands. The research was overseen by Rachel Greenstadt, CSE professor and faculty member of the NYU Center for Cybersecurity and Gunes Acar, associate professor at Radboud University.

AdLens identified three major categories of deceptive advertising:

• Scareware Ads: Ads with fake warnings include those claiming that users' devices were infected, hacked or compromised generated more than 16 million impressions. The researchers identified 238 such ads within the 189k ads examined.
• Misleading software promises: The researchers identified 3,346 ads making false or unrealistic claims about what software could do, including recovering photos deleted years earlier. These ads generated nearly 90 million impressions.
• Deceptive designs: The researchers identified 258 ads that concealed the advertiser's identity or what was being promoted, sometimes displaying little more than a button labeled "Continue" or "Access Now." One such ad generated 2.1 million impressions.

One advertisement promising to reveal anyone's GPS location using only a phone number generated 14.5 million impressions and remained active for more than two years.

“These ads are only the tip of the iceberg, as the ad transparency platform hosts more than two billion ads of which only a small portion was analysed,” said Roongta. “The high incident of deception within this small sample itself highlights the magnitude of these scams being propagated to the users.”

The researchers also examined apps promoted through deceptive advertisements. Of 147 identified apps, 34 were no longer available in app stores even though their advertisements remained active.

AdLens uses publicly available AI models (small language models or SLMs) to identify suspicious advertising efficiently. It first searches for language resembling known deceptive tactics, then examines the text and images of likely offenders. By concentrating intensive analysis on a smaller group of ads, the system reduces the computing resources needed for large-scale investigations.

The study builds on earlier research by Roongta and Greenstadt, which found that users of Adblock Plus's "Acceptable Ads" feature encountered 13.6% more problematic advertisements than users browsing without ad blockers. Together, the studies reveal shortcomings in systems intended to improve the online advertising experience.

The researchers have made AdLens publicly available so regulators, consumer protection organizations and independent researchers can investigate advertising practices without relying exclusively on platforms' own assessments. Although this study focused on Google, the technology could be adapted to examine other advertising platforms, including Meta and TikTok.

The researchers caution that ad impressions measure exposure, not confirmed financial losses or malware infections. Nevertheless, their findings demonstrate how deceptive advertising can persist despite existing safeguards and underscore the need for stronger enforcement and independent oversight.

In addition to Roongta and Greenstadt, the authors of the AdLens paper are Marwan Adnan Darwish, Masoud Poorghaffar Aghdam, and Gunes Acar, all of Radboud University. The research was supported by the National Science Foundation and the Netherlands Organisation for Scientific Research.

Roongta, R., Darwish, M., Aghdam, M., Greenstadt, R., & Acar, G. (n.d.). Deceptive software ads. https://doi.org/10.1145/3830454.3846781