Google AI Threat Defense turns Gemini loose on vulnerability triage

Google Cloud rolled out AI Threat Defense this morning. The pitch is blunt: fight AI with AI.

The platform stitches together three assets Google spent billions acquiring. Mandiant brings frontline incident response data. Wiz, the $32 billion cloud security pickup from late 2024, maps attack surfaces across multi-cloud environments. Gemini does the reasoning. A tool called CodeMender generates and tests patches, then drops them directly into a developer’s IDE or CLI at build time.

The architecture follows a four-step loop. First, map everything: exposed APIs, applications, configurations, identities, permissions. Second, run AI-driven adversarial testing against internet-facing applications and critical systems. Third, autonomously generate and deploy verified fixes. Fourth, machine-speed detection and response with outcome tracking.

The interesting part is step three. Google says AI Threat Defense reduces time-to-remediate from weeks to minutes. The system generates patches, tests them, tags libraries across source control and production, and coordinates rollouts. Each patch gets verified before deployment. The tool rewrites older code to memory-safe languages and analyzes library dependency chains.

Google’s framing positions this as a response to attackers who already use AI to accelerate their operations. The blog post is explicit: traditional attack surface management tells you what is exposed. Organizations now need “an AI penetration tester that can continuously analyze every exposure, determine whether it can actually be exploited, and understand what it would enable an attacker to do before attackers do the same.”

This is also a competitive move. Anthropic launched Mythos, its security-focused Claude variant, earlier this year. OpenAI has Daybreak. Google is building the same play but stacking acquired capabilities instead of growing them organically.

The Mandiant intelligence layer matters here. Incident response data from real breaches feeds the prioritization engine. That is a different data moat than what Anthropic or OpenAI can offer. Wiz’s cloud security footprint gives Google visibility into multi-cloud environments that pure-play AI companies cannot match.

Google says the platform is available now. Pricing was not disclosed.

Sources: SecurityWeek, CryptoBriefing, Indian Express, CSR Journal, Google Cloud blog