Google’s Agentic Defense Puts Its $32 Billion Wiz Bet to Work

Google has folded its $32 billion acquisition of Wiz into Agentic Defense, a lineup of AI security agents built to catch attackers moving at machine speed. The company is selling the launch as a shield against AI-powered hacking. Its own threat researchers are telling a more complicated story.

Google’s newly published M-Trends 2026 report, drawn from more than 500,000 hours of Mandiant incident response work, found that attackers now hand off a compromised network to a second crew in a median of just 22 seconds, down from over eight hours in 2022. The same report states plainly that most 2025 breaches still trace back to human and systemic failures, not artificial intelligence.

A $32 Billion Bet Finally Closes

Google first approached Wiz in July 2024 with an offer of $23 billion. Wiz turned it down. Cofounder and chief executive Assaf Rappaport bet the company could grow into something bigger and pursue an initial public offering instead.

Less than a year later, he was proven right. Talks resumed in early 2025, and in March, Google returned with a cash offer of $32 billion, laid out in a securities filing disclosing the merger agreement. Wiz accepted.

The deal still needed sign-off from regulators on two continents. The Department of Justice, which had opened an antitrust review into whether the acquisition would weaken competition in cybersecurity, issued an early termination of that investigation in November 2025. European Union regulators cleared it in February 2026. Google closed the purchase on March 11, 2026, almost exactly a year after the agreement was signed.

It stands as Google’s largest acquisition ever and one of three megadeals that have redrawn the cybersecurity map since 2024.

Industry trackers describe the Wiz purchase as the largest all-cash cybersecurity acquisition on record, ahead of both rivals in the table above.

Five Agents Built for Machine-Speed Defense

Agentic Defense sits on top of Google Security Operations and folds in Wiz’s scanning and remediation tools. An existing Triage and Investigation agent, already in use, is joined by three new agents entering preview: one hunts for novel attack patterns, one closes detection gaps, and one pulls in outside threat context.

• Triage and Investigation agent – already available, sorts and investigates incoming security alerts
• Threat Hunting agent – new, proactively searches for stealthy attack patterns that slip past standard detections
• Detection Engineering agent – new, finds coverage gaps and writes fresh detection rules for specific threats
• Third-Party Context agent – new, enriches analyst workflows with data pulled from outside sources
• Wiz Security Agents – Wiz’s own agents, built to identify and respond to cloud risk at speed

Google also made remote Model Context Protocol server support generally available inside Security Operations, with direct access through its chat interface still in preview. A new dark web intelligence feed for Google Threat Intelligence, also in preview, tested at 98% accuracy in the company’s internal evaluations.

Francis DeSouza, Google Cloud’s chief operating officer and president of security products, said the moment calls for a different kind of defense: “The AI era demands a new security era.” He argued Google holds an edge rivals cannot easily copy, pointing to its own chips, cloud and models: “We’re the only security company that has our own AI stack.”

Google detailed the buildout in a post introducing Agent Identity and Agent Gateway for enterprise agents, new pieces of its Gemini Enterprise Agent Platform that give AI agents their own credentials and enforce policy on agent-to-agent traffic.

Attackers Now Hand Off Access in 22 Seconds

Google published its M-Trends 2026 report the same month it began rolling out Agentic Defense, and the numbers explain the urgency better than the product pitch does. Mandiant, Google’s incident response arm, drew its findings from more than 500,000 hours of investigations conducted in 2025.

The report’s sharpest figure covers what Mandiant calls the initial access handoff, the gap between an intruder first breaking in and passing control to a second crew that does the actual damage. In 2022, that median gap ran past eight hours. In 2025, it fell to 22 seconds, a drop of more than 99.9%. Mandiant said the pattern reflects tighter coordination between the specialists who break in and the specialists who cash out, with access brokers increasingly delivering malware directly on a partner’s behalf instead of selling it on underground forums.

• 22 seconds – new median handoff time between initial access and a second attack crew, down from over eight hours in 2022
• 9% – share of 2025 investigations that showed this division-of-labor pattern, up from 4% in 2022
• 13% – share of 2025 investigations tied to ransomware, with operators destroying backups to force payment
• 714 – new malware families Google’s Threat Intelligence Group identified in 2025, up from 632 in 2024

The executive summary of the 2025 findings also found dwell time, the days an attacker sits undetected inside a network, ticked up to a median of 14 days in 2025 from 11 in 2024, though still far below the 146 days Mandiant measured a decade ago.

Human Error Still Drives Most Breaches, Google Says

Agentic Defense is being pitched as a way to counter attackers who use generative AI to move faster and hide better. In the post announcing the report, Google’s threat intelligence team downplayed AI’s role in 2025 breaches, writing:

M-Trends 2026 confirms attackers are abusing AI within compromised environments, however, we do not consider 2025 to be the year where breaches were the direct result of AI. The vast majority of successful intrusions still stem from fundamental human and systemic failures.

Google’s Threat Intelligence Group wrote that in the M-Trends 2026 announcement, separating AI as a tool attackers use once inside a network from AI as the reason they got in.

The rest of the report backs that up. Exploits remained the leading initial access vector for the sixth consecutive year, present in 32% of investigations where Mandiant could identify one. Voice phishing, a human con that runs on a live phone call rather than an algorithm, climbed to second place at 11%. Email phishing, the technique most associated with automated, AI-written lures, kept shrinking as a share of cases.

Why Does Wiz Still Serve Google’s Cloud Rivals?

Wiz continues to support Databricks, AWS Agentcore, Microsoft Azure Copilot Studio and Salesforce Agentforce even under Google’s ownership, extending the same cross-cloud coverage it sold before the acquisition. That matters because the Justice Department’s antitrust review specifically examined whether Google owning Wiz would weaken security competition across rival clouds.

Wiz built its reputation by watching accounts inside Amazon Web Services and Microsoft Azure as closely as inside Google Cloud, correlating assets, identities and vulnerabilities into what the company calls its Security Graph. Keeping that reach intact after an acquisition of this size was not guaranteed.

Alphabet is already living with a separate EU order opening Android’s AI and search data to rivals by 2027. The Wiz review carried a narrower question: whether folding a cross-cloud security watchdog into the search giant would blunt competition among cloud providers.

Google’s answer, so far, is to keep expanding Wiz outward rather than pulling it inward. At Cloud Next, Wiz added support for Databricks and for agent-building platforms across AWS, Microsoft and Salesforce, plus integrations with Cloudflare’s AI security tools and the Vercel platform.

Cybersecurity Consolidates Around a Few Giants

The Wiz deal is one entry in a broader wave. Enterprise security teams have grown tired of stitching together dozens of point tools, and buyers increasingly favor platforms that unify detection across endpoints, networks, identities and clouds.

CrowdStrike spent much of the past two years rebuilding customer trust after a July 2024 outage that took down Windows systems worldwide. Palo Alto Networks has pursued its own platform strategy through a string of acquisitions. The logic in both cases mirrors Google’s: a vendor that sees telemetry from endpoints, cloud workloads and identity systems in one place can train AI models no single-purpose tool can match.

That logic extends past the hyperscale clouds. Banks in India are already rewriting their cyber threat models around AI-driven identity fraud and real-time payments, evidence the shift toward AI-native defense reaches well beyond cloud providers.

Three of the five Security Operations agents Google announced remain in preview, with no stated date for general availability. The dark web intelligence feed and direct chat access to Google’s threat data carry the same preview label. For now, Google is asking enterprise customers to trust a $32 billion bet on speed before all of its pieces are finished.

Frequently Asked Questions

What Is Wiz’s Security Graph?

Wiz’s Security Graph is the graph-based analysis engine that made the company one of the fastest-growing names in cloud security after its founding in 2020. It correlates cloud assets, identities, vulnerabilities and exposures across multiple providers into a single map, which is what let Wiz watch Amazon’s and Microsoft’s clouds as closely as Google’s even before the acquisition.

What Does CNAPP Mean, and Is Wiz One?

CNAPP stands for cloud-native application protection platform, an industry category that bundles vulnerability scanning, posture management and workload protection into one tool instead of several. Wiz is widely categorized as a CNAPP, which is why Palo Alto Networks’s Prisma Cloud and CrowdStrike’s Falcon Cloud Security compete directly with the product Google now owns.

How Big Is Mandiant’s M-Trends Report?

The M-Trends report has run annually for 17 years, according to Google, and the 2026 edition is built from more than 500,000 hours of incident response investigations conducted in 2025. Google publishes it through its Threat Intelligence Group, which now includes Mandiant.

Which Sectors Do Attackers Target Most?

High-tech companies were the most targeted sector in Mandiant’s 2025 caseload, followed by financial services, business services and healthcare. Roughly 30% of the incidents Mandiant investigated were financially motivated, and 40% involved data theft rather than disruption alone.

What Is the Model Context Protocol Google Added Support For?

Model Context Protocol is a standard that lets AI agents and outside tools share context and take action across systems without custom, one-off integrations. Google made remote server support for it generally available inside Security Operations, letting external tools plug into its detection agents directly, while access through its own chat interface remains in preview.