Google Cloud rolled out AI Threat Defense this week, an enterprise security platform that hunts for code vulnerabilities, writes patches, and ships them with limited human input. The company laid it out on its security blog. The pitch: close gaps before attackers, who are also using AI to find flaws faster, can reach them.
Most of the parts aren't new. Wiz, which Google bought for $32 billion, maps exposed apps, APIs, and identities, then runs a red-team agent to figure out which holes are actually exploitable instead of theoretical. That last bit matters, since the usual problem isn't finding flaws but drowning in them.
For scanning, Google runs several models at once. Lighter ones do broad, continuous coverage; frontier models get reserved for the highest-risk assets. Patching falls to CodeMender, the DeepMind agent introduced on October 6, 2025. It steps into the IDE, swaps out vulnerable code, and can rewrite older codebases into memory-safe languages. Before any fix ships, the platform generates tests to check it, and tags which model wrote which patch.
Google calls the workflow "autonomy under human supervision," though autonomous remediation carries an obvious risk: an automated fix that breaks production. Launch partners include Accenture, Deloitte, and PwC. The Monitor stage hands off to agents from Google Security Operations for active threat hunting.
Bottom Line
Google wired together Wiz ($32B acquisition), DeepMind's CodeMender, Gemini, and Mandiant into one platform that auto-patches code under human review.
Quick Facts
- Platform name: Google AI Threat Defense
- Wiz acquisition: $32 billion
- CodeMender introduced October 6, 2025 by DeepMind
- Four-stage framework: Prepare, Scan and Prioritize, Remediate, Monitor
- Launch partners: Accenture, Deloitte, Netenrich, PwC, TENEX.AI
