Detecting and containing AI-powered threats with Google Security Operations agents

Google argues that defenders cannot rely on patching to stay safe, because attackers now move faster than fix cycles. Citing its M-Trends 2026 research, the company says the exploitation of vulnerabilities has become the most common way attackers first break in. The report estimates a mean time to exploit of minus seven days, meaning attacks often start before a patch is even released.

The numbers from the 2026 Verizon Data Breach Investigations Report reinforce the point. Across more than 13,000 organizations studied, only 26% of vulnerabilities on the CISA Known Exploited Vulnerabilities list had been fully remediated, and the median time to fully patch after detection was 43 days. Google calls this stretch the remediation gap and positions continuous monitoring as the way to cover it.

Google describes these capabilities as running at machine scale and machine speed. Paired with AI Threat Defense, the goal is a platform that responds faster than human teams working alone, so defenders can keep pace with attacks that are themselves accelerated by AI.

The Detection Engineering agent, available in preview, automatically translates new exploitation patterns of unpatched flaws into detection rules tailored to a specific environment. It pulls from a wide range of inputs, including Google Threat Intelligence, attack patterns curated by Mandiant, offensive tool repositories, red and purple team reports, autonomous malware analysis, open-source detection blogs, and a company's own security telemetry.

Rather than wait for an exploit to land, the agent proactively builds new rules and tests them with synthetic events to confirm coverage ahead of time. Google illustrates this with the Axios supply chain attack, where the agent flagged execution phases, spotted blind spots at the entry and exit points, and generated custom YARA-L rules to close the gaps.

The Triage and Investigation agent autonomously digs into alerts, gathers supporting evidence, and returns a verdict with a full explanation. Google says it reduces a typical 30-minute manual analysis to 60 seconds using Gemini, and reports that it has already investigated more than 5 million alerts.

By automating that decision-making, the agent frees analysts to focus on harder problems while routine alert closure and remediation steps move faster. This is the piece of the lineup that is generally available rather than in preview.

The agentic automation capability, offered in preview, contains active attacks by combining dynamic AI agents with deterministic enterprise playbooks. This hybrid design lets AI handle decision-making and remediation workflows while analysts retain control over critical, high-impact actions. The intent is speed without handing every consequential decision to software.

The Threat Hunting agent, also in preview, proactively looks for novel attack patterns and quiet adversary behavior that bypasses traditional tools. It scours petabytes of enterprise telemetry, including older logs, for subtle anomalies. Google says this shifts a security operations center from a reactive stance to a more proactive one, catching activity that earlier rules missed.

For a business that runs software it did not build and cannot always patch, the message is that monitoring and fast response now matter as much as fixing flaws. The agents aim to shrink the window between an attack starting and someone noticing, which is where much of the damage happens.

It is worth noting that most of these agents are in preview, an early-access stage, so capabilities and results described by Google reflect its own examples rather than independent benchmarks. The figures on time saved and breach reduction come from Google and the reports it cites, not from third-party testing of the agents.