Google says its systems caught more than 99% of policy-violating ads before they appeared in 2025, yet researchers identified thousands of deceptive software ads that ran, some for years

1

.

The finding comes from researchers at NYU Tandon and Radboud University, who built an AI tool called AdLens and pointed it at Google's Ads Transparency Center, the company's public archive of ads

2

. They analyzed about 189,000 software-related ads collected in March 2026 and identified thousands of deceptive ones, which together drew over 100 million impressions across European markets

2

,

1

.

What the ads promised

The deceptive ads fell into three groups. The study counted 238 scareware ads, the fake infection and hacking warnings, with over 16 million impressions. It tallied 3,346 ads making false or unrealistic software claims, such as recovering long-deleted photos, with nearly 90 million impressions. And it picked out 258 ads using deceptive designs that hid who the advertiser was, sometimes showing nothing but a "Continue" or "Access Now" button

2

.

One ad promised to pinpoint anyone's location from just a phone number. It ran for more than two years and drew 14.5 million impressions

2

,

1

. Some ads stayed up for years, and others kept running even after being reported to Google

2

. Of 147 apps promoted through the deceptive ads, 34 had been pulled from app stores while their ads stayed active

2

.

Testing Google's reporting pipeline

The researchers also reported 24 ads through Google's public reporting tool. Of those, 15 were judged "no violation" and 6 were removed

1

. And when Google removed one ad tied to the TamperedChef malware campaign, 41 other ads pointing to the same site stayed live

2

,

1

.

Per Google's own 2025 Ads Safety Report, 8.3 billion ads were blocked or removed, including 602 million related to scams

1

. Lead author Ritik Roongta, an NYU Tandon computer science and engineering PhD student, put the gap plainly: "Google has systems in place to screen advertisements, but we found thousands of deceptive ads that still reached users, sometimes for years"

2

.

The researchers name limits on what the study can claim. An impression count shows exposure only, not confirmed losses or infections, and the sample covered just a small portion of the more than two billion ads on the platform. Roongta calls these ads "the tip of the iceberg"

2

,

1

.

The findings land as regulators press the industry. In July, UK regulator Ofcom recommended rules under the Online Safety Act requiring platforms including Google to ban fraudulent advertisers, with potential fines up to 10% of global revenue

1

.

AdLens uses small public AI models to flag language resembling known deceptive tactics before examining likely offenders, cutting the computing cost of large-scale screening. It has been made publicly available for regulators, consumer groups and researchers, and could be adapted to platforms like Meta or TikTok

2

. The study will be presented November 18 at the ACM Conference on Computer and Communications Security in The Hague

2

. "We shouldn't have to rely entirely on advertising platforms to tell us whether their safeguards are working," Roongta said

2

.

© 2026 TheOutpost.AI All rights reserved