Google Uses Gemini to Scan Dark Web: Processes Nearly 10 Million Posts Daily with 98% Accuracy

Google launched a dark web intelligence service in Google Threat Intelligence, using AI agents powered by the Gemini model to crawl and screen 8 to 10 million dark web posts daily, identifying threats related to specific organizations with an internal test accuracy of 98%. After customers verify their identity, Gemini automatically generates organizational profiles (covering business operations, VIPs, brands, tech stacks, etc.) within minutes and produces categorized threat alerts by retrieving 7 days of data.

Traditional dark web monitoring tools rely mainly on keywords and regex matching, with a false positive rate as high as 80% to 90%. Google also added an AI agent (preview version) in Google Security Operations that can independently investigate alerts and collect evidence, and supports enterprises to build custom security agents via remote MCP servers (officially launched). Gemini integrates knowledge from Google Threat Intelligence Group's human analysts, who track 627 threat organizations.