By Every major cybersecurity vendor launched an AI product this week at RSAC 2026 in San Francisco. CrowdStrike, Palo Alto Networks, Cisco, Google (with its freshly acquired Wiz), Arctic Wolf, Dell, Rubrik. All on Day 1. The trade press covered it as a parade of product launches. Dozens of articles, each dutifully listing features and capabilities.
That framing misses what happened.
The real story is not that security companies are adding AI features. The real story is that the entire security industry just restructured itself around a single architectural assumption: that the things companies need to protect are no longer humans sitting at keyboards, but autonomous software agents acting on their behalf. And every major vendor reached that conclusion at the same time. That kind of synchronized pivot does not happen because marketing departments coordinated. It happens because the underlying economics changed.
The trigger nobody is talking about
Start with a number. According to ETR data presented at RSAC this week, 37% of organizations already have AI agents deployed or in active testing. That is up 10 percentage points from last year. But here is the part that should worry you: only 3% of those organizations say they have broad security controls in place for those agents. Twenty percent admit they have no agent-specific controls at all.
That gap, between deployment speed and governance readiness, is the structural force driving every announcement on the Moscone floor this week. Vendors are not launching AI products because AI is trendy. They are launching because their customers just created a massive new attack surface and have almost nothing protecting it.
The shift also showed up in spending patterns. For the first time, LLM and generative AI protection surpassed cloud security as the top area where enterprises plan to increase spending, according to ETR's annual security study. Cloud security had held that position for two years running.
What the vendors actually built
The specifics matter because they reveal a shared diagnosis.
CrowdStrike expanded its Falcon platform into what amounts to an AI surveillance layer. A new feature called Shadow AI Discovery identifies AI applications, agents, LLM runtimes, and Model Context Protocol servers running across devices. EDR AI Runtime Protection tracks how AI applications behave on endpoints, logging commands, scripts, file activity, and network connections. The company is also watching AI agents inside SaaS platforms like Salesforce Agentforce, Microsoft Copilot, and ChatGPT Enterprise. CrowdStrike's thesis is blunt: the endpoint is where AI actions execute, so the endpoint is the control plane.
Palo Alto Networks went a different direction. Prisma AIRS 3.0 secures the full lifecycle of agentic AI, from discovery through runtime protection. The company also launched Prisma Browser for Business, which embeds large language models directly into the browser with inline data loss prevention. If an agent issues an unauthorized action, the system figures out whether a human prompt or an autonomous agent caused it.
Google used the conference to formally confirm the completion of its $32 billion Wiz acquisition, the largest purchase of a private venture-backed U.S. company in history. Wiz immediately launched its AI Application Protection Platform (AI-APP), designed to secure every layer of AI applications from infrastructure to agents to runtime. Meanwhile, Google introduced what it calls an "agentic SOC," powered by Gemini models, where AI agents autonomously investigate alerts, gather evidence, and deliver verdicts. Google's Mandiant unit released data showing attackers now reduce the window for defenders to respond to as little as 22 seconds.
Cisco extended zero-trust identity controls and AI Defense capabilities to cover non-human identities and autonomous AI agent workflows, integrating Splunk SOC automation into the mix.
Who benefits, who loses
The pattern across all these launches points to one conclusion: the security industry is consolidating around platforms, and AI is the accelerant.
ETR's data backs this up. The "best-of-breed" rationale for adding new security suppliers hit a two-year low. Only 5% of respondents plan to decrease security spending, but the money is flowing to fewer vendors. Niche startups still enter where platforms fall short, but the most likely outcome is acquisition and absorption.
The winners are the companies large enough to offer an integrated platform that covers the full AI attack surface: endpoint, cloud, SaaS, identity, and runtime. CrowdStrike, Palo Alto Networks, and Google (with Wiz) are positioning for exactly that.
The losers are point-solution vendors who built products for a world where the things being protected were predictable software systems operated by humans. That world is being replaced. Autonomous agents now carry credentials, make API calls, access sensitive data, and operate with elevated privileges. No human in the loop.
As Caleb Sima of the Cloud Security Alliance put it at the conference's opening session, the existing model of detection and response is fundamentally broken at machine speed. It takes months to create new detections and months more to identify breaches. When adversaries deploy parallel agents that scan, exploit, and move laterally in minutes, human-driven security processes simply cannot keep pace.
The historical parallel
This has happened before. In the mid-2000s, the shift from on-premises infrastructure to cloud computing created a similar structural break. The companies that recognized cloud was not just a deployment option but a fundamentally different security surface, Palo Alto Networks among them, became the dominant players of the next decade. The companies that treated cloud as a feature to bolt onto existing products got acquired or faded.
The same pattern is playing out now, faster. Cloud took roughly a decade to force the security industry's hand. AI agents are doing it in under two years. The 37% deployment figure from ETR is not a forecast. It is current state. And the governance gap (3% with broad controls) means the window for vendors to establish themselves as the default AI security platform is open right now but closing fast.
What this looks like in five years
The cybersecurity industry will not have dozens of AI security startups in 2031. It will have three or four platforms that own the AI security layer the way CrowdStrike owns endpoint detection today. Google's $32 billion bet on Wiz is not a cloud security play. It is a bet that whoever controls the AI security platform controls the next era of enterprise infrastructure spending.
The vendor that wins will be the one that solves the identity problem, because the core challenge is not detecting threats to AI systems. It is governing what AI agents are allowed to do. Every product launched this week, from CrowdStrike's shadow AI discovery to Cisco's zero-trust for non-human identities to Palo Alto's agent lifecycle management, is a different approach to the same question: who is this agent, what can it access, and should we let it?
The companies answering that question best in the next 18 months will own the security market for the next decade. The rest will become acquisition targets.
Jules Okonkwo covers technology for The Daily Vibe.
