Agentic Defense

Google Cloud's AI-era cybersecurity framing: use AI agents, threat intelligence, security operations, and Wiz cloud/AI security context to prevent, detect, investigate, and remediate threats across code, cloud, runtime, and AI systems.

Key points

• The keynote says AI expands the attack surface because enterprises are deploying agents that access sensitive data while adversaries use AI to attack faster ^[src-044].
• Google Cloud combines Google Threat Intelligence and Security Operations with Wiz's Cloud and AI Security Platform for agentic defense ^[src-044].
• Triage and Investigation Agent has processed more than 5 million alerts, reducing a typical 30-minute manual analysis to 60 seconds ^[src-044].
• New agents include Dark Web Intelligence, Threat Hunting Agent, Detection Engineering Agent, Wiz Red Agent, Wiz Blue Agent, Wiz Green Agent, and Wiz Agentic Workflows ^[src-044].
• Wiz AI Application Protection Platform secures AI applications across infrastructure, data, access, models, agents, and applications from code to runtime ^[src-044].
• Google Cloud Fraud Defense extends reCAPTCHA toward discerning the legitimacy of bots, humans, and agents ^[src-044].

Related concepts

• Enterprise Agent Governance
• Governance Observability
• Agent Forensics
• Agent Circuit Breakers
• Agentic Enterprise

Source references

• ^[src-044] Thomas Kurian — "Welcome to Google Cloud Next '26" (2026-04-22)

Evidence checked on 14 September 2026

Derek Abdine argues that effective remediation needs endpoint-specific reasoning before, during and after a fix: services, disk space, configuration state and ownership change what is safe to apply. ^[src-444] Auto-generated captions can misrecognize names, technical terms and numbers. Interview with a remediation vendor CEO; product effectiveness and exploit-window claims were not independently benchmarked.

• Autonomous AI Remediation: Solving the Vulnerability Backlog | Derek Abdine, Furl

OpenAI September 3 announcement commits $1B globally to subsidized Daybreak access, training, technical support and partnerships for essential-service defenders. ^[src-458] Commitment of subsidized services/support, not cash grants or confirmed spending. Do not infer unrestricted Astra access for recipients; reported access was phased and restricted. Allocations, uptake and outcomes not evaluated.

• OpenAI commits $1B in AI credits to frontline cyber defenders

Jyn argues that increasingly accessible cyber-capable models shorten the time available to harden software and that deploying patches is the remaining operational bottleneck. ^[src-460] The one-year timeline is a forecast, not an established deadline or measured probability.

• We have a year to fix security everywhere