Google Scans Dark Web with Gemini: Processes Nearly 10 Million Posts Per Day, 98% Accuracy, Traditional Tools Have a 90% False Positive Rate

According to 1M AI News, Google has launched a dark web intelligence service in Google Threat Intelligence, using AI agents based on the Gemini model to crawl and screen 8 to 10 million dark web posts daily, identifying threats related to specific organizations, achieving an internal test accuracy of 98%. Once the client confirms their identity, Gemini automatically generates an organizational profile within minutes (covering business operations, VIPs, brands, tech stack, etc.) and retroactively generates classified threat alerts for the past 7 days. Traditional dark web monitoring tools mainly rely on keywords and regex matching, with a false positive rate as high as 80% to 90%.

Google has simultaneously added an AI agent (in preview) to Google Security Operations that enables autonomous investigation of alerts and evidence collection, and supports enterprises in building custom security agents through remote MCP servers (now officially launched). Gemini has also integrated the knowledge of human analysts from the Google Threat Intelligence Group, a team that tracks 627 threat organizations.