Google turns Gemini loose on malvertising as attackers scale up with generative AI

Google is leaning harder on Gemini to police its ad network, citing 8.3 billion ads blocked or removed and 24.9 million advertiser accounts suspended in 2025, with 602 million of those ads tied to scams. The shift reflects an arms race: criminals are using generative AI to mass-produce convincing ads that impersonate legitimate brands, cloak their true destinations, and funnel victims into malware, credential theft, and wallet-draining crypto sites.

The technical change is a move away from keyword-centric detection toward signal-rich classification. Gemini evaluates advertiser behavior, account history, campaign patterns, and inferred intent across billions of signals to flag ads at submission rather than post-click. By late 2025, the majority of Responsive Search Ads were being reviewed instantly, with harmful content blocked before it ever served. Google plans to extend that submission-time enforcement to more ad formats.

The secondary claim worth noting is an 80% reduction in incorrect advertiser suspensions, which matters because false positives have historically been the main cost of aggressive automated enforcement. Whether Gemini’s pattern-based approach holds up against adversarial advertisers — who are themselves using generative models to vary creative, infrastructure, and personas — is the real test, and one Google’s own numbers don’t yet answer.