In a digital age where online safety claims are increasingly weaponized for political control, a new study reveals that Google’s advertising safeguards are failing on their own terms. Researchers at New York University’s Tandon School of Engineering and Radboud University in the Netherlands built an AI tool called AdLens and analyzed 189,000 software-related ads from Google’s public archive. From early 2023 to March 2026, they identified thousands of deceptive ads that collectively generated more than 100 million impressions across Europe. These results expose how automated systems, even those run by a trillion-dollar company, cannot be trusted to police the content they profit from.
Researchers classified the deceptive ads into three categories. The most alarming was scareware, which uses fake virus warnings to pressure users. One Polish-themed ad claimed “178 viruses were found on your Android phone” and was shown 1.1 million times. The 238 scareware ads found collectively generated more than 16 million impressions.
The second category involved false advertising promises, such as apps claiming to recover photos deleted years earlier or reveal any phone’s GPS location from a number alone. These 3,346 ads drew nearly 90 million impressions. One such phone-tracking ad remained active for over two years, generating 14.5 million impressions. The third group used misleading designs, hiding advertiser identity behind buttons like “Continue” that led to paid subscriptions.
Of 147 apps promoted by flagged ads, 34 had already been removed from app stores while their ads kept running. The researchers also tested Google’s “Report this ad” button. When they reported more than 20 ads, Google judged most as not violating its rules. In a critical case, Google removed one ad tied to a known malware campaign called TamperedChef, but 41 other ads directing users to the same website remained active.
This pattern of partial enforcement mirrors broader concerns about self-regulation in the tech industry. Google claims its systems catch over 99% of policy-violating ads before they appear, yet the study reveals that even direct human reporting cannot stop the remaining fraction. This raises fundamental questions about whether these platforms can be trusted to police themselves.
The findings come amid increasing government efforts to regulate online content worldwide. New York State has pursued its own Online Hate Speech Law and considered bills targeting so-called “misinformation.” In the U.K., the Online Safety Bill threatens tech firms with massive fines for failing to enforce censorship rules, while the European Union’s Digital Services Act mandates independent audits and crisis response mechanisms for large platforms.
But this study reveals a paradox. Governments demanding stricter content moderation rarely scrutinize the effectiveness of existing systems. The U.K.’s communications regulator, Ofcom, has recently recommended new rules under the Online Safety Act that would force Google to ban fraudulent advertisers or face fines of up to 10% of global revenue. However, Ofcom’s own leadership includes former Google and Meta executives, raising questions about regulatory capture.
The ACLens tool used in this study relied on publicly available AI models and cost minimal computing resources. The researchers have made it publicly available so consumer protection organizations and independent regulators can investigate advertising platforms without depending on the platforms’ own reports. Ritik Roongta, the lead author, noted that “we shouldn’t be relying entirely on the advertising platforms to tell us whether their protections are working.”
The study measured ad impressions, not confirmed financial losses or malware infections. Yet the scale of the problem is undeniable. A system that blocks billions of ads annually but still lets a reported scareware ad stay up has a gap worth closing. As governments rush to impose new censorship mandates on Big Tech, this research reminds Americans that corporate promises of safety systems are not guarantees of truth. The most reliable protection remains an informed public and robust independent oversight, not centralized control by either private monopolies or government regulators.
Sources for this article include: