Cybersecurity — 2026-04-19
Google 2025 Ads Safety Report: Gemini-powered screening blocked record 8.3B policy-violating ads, 24.9M accounts suspended
Google published its 2025 Ads Safety Report Thursday disclosing 8.3 billion ads blocked, up roughly 63% from 5.1B in 2024, and 24.9M advertiser accounts suspended, with more than 99% of policy violations stopped pre-serve. Gemini-powered pattern detection is credited for the step-change. The model analyzes account age, behavior patterns, and campaign activity across hundreds of billions of signals to identify fraudulent intent earlier than rule-based systems. Specific scam-ad categories saw 602M removals and 4M account suspensions. The report is a data point in the debate over LLM-enabled trust-and-safety at scale.
Analysis
Within the coming year, the 8.3B blocked-ad figure is likely consistent with either genuine Gemini detection gains or a step-change in adversarial LLM-assisted ad generation, and Google's self-reported numbers, the sole source absent external validation, cannot distinguish between the two. The 602M scam-ad removals and 4M associated account suspensions are the more actionable disclosures: they benchmark the current base-rate of LLM-assisted scam fraud clearing Google's serving tier. The 99% pre-serve block rate carries a structural blind spot: it counts nothing of adversarial advertisers who migrated to Meta or TikTok rather than testing Google inventory, a substitution effect no single-platform disclosure can reach.
Within the coming year, the 8.3B blocked-ad figure is likely consistent with either genuine Gemini detection gains or a step-change in adversarial LLM-assisted ad generation, and Google's self-reported numbers, the sole source absent external validation, cannot distinguish between the two. The 602M scam-ad removals and 4M associated account suspensions are the more actionable disclosures: they benchmark the current base-rate of LLM-assisted scam fraud clearing Google's serving tier. The 99% pre-serve block rate carries a structural blind spot: it counts nothing of adversarial advertisers who migrated to Meta or TikTok rather than testing Google inventory, a substitution effect no single-platform disclosure can reach.
1 sources
- Google wipes out 602 million scam ads with Gemini on duty -
Help Net Security