Our fight against fraud: 5 ways we’re keeping you safer

Source: Google Date: 2026-05-13 URL: https://blog.google/innovation-and-ai/technology/safety-security/scams-fraud-protection/

Summary

Google outlines five fraud-defense capabilities: ML-powered blocking of ~15 billion unwanted Gmail messages daily and 99%+ policy-violating ads in 2025; user-facing tools including Security Checkup (passkeys, 2-step verification) and Circle to Search scam assessment on Android; the Be Scam Ready educational program backed by $5M in Google.org grants; participation in the Global Signal Exchange, a shared threat-intelligence clearinghouse holding 1.2 billion signals; and direct law enforcement collaboration targeting networks like Lighthouse and BadBox. Chrome’s real-time dangerous-site prediction is called out as an AI-powered client-side defense.

Implications

• Mythos / Project Glasswing thread. Google’s at-scale fraud detection (15B emails/day, 99%+ ad policy enforcement) is built on the same ML stack that underpins their safety positioning. The Global Signal Exchange is structurally analogous to the Glasswing disclosure program — both treat threat intelligence as a shared resource rather than a competitive differentiator, which changes the security economics for smaller players.
• Claude Code security surface thread. The Circle to Search scam-assessment feature on Android is a consumer-facing precedent for on-device AI safety tooling. As agentic systems proliferate, the same pattern (local AI evaluates content before user acts) is the architecture needed for agent permission prompts — Google is building user intuitions for this interaction model at scale.
• MCP at enterprise scale thread. The Global Signal Exchange’s 1.2 billion signals is a data-layer moat for fraud detection. Enterprises evaluating agentic platforms will increasingly weight integrated threat intelligence as a deployment requirement, not a vendor add-on.