At Fal.Con 2026 in Las Vegas, CrowdStrike and OpenAI announced an expanded partnership that goes straight at one of the most urgent problems in enterprise AI: nobody knows how many AI agents are actually running inside their business.

The answer, at least for one unnamed Fortune 500 company, turned out to be 18,000. Only 300 had been approved.

What Was Announced

The partnership has two parts. First, CrowdStrike is extending enterprise security to OpenAI’s Codex agents. Second, OpenAI’s GPT-5.6 Cyber model is coming to the Falcon platform to power faster threat detection and prioritisation.

The centrepiece is Falcon Guardian, a new AI Detection and Response (AIDR) solution that monitors AI agents at runtime, on the endpoint where they actually execute, not just at the governance or policy level.

What makes this different from existing AI governance tools is that it requires no hooks or SDK. The Falcon sensor already deployed across enterprise endpoints now extends automatically to AI agents, fusing AI telemetry and OS telemetry from the moment a prompt is sent through to every process the agent touches.

On day one, the Fortune 500 example, Falcon Guardian discovered agents from OpenAI Codex, Claude Code, Cursor, and Kiro running across their environment, along with a pile of unclassified shadow AI that no one in IT had sanctioned.

Why This Matters

The agent sprawl problem has been building since early 2026. Developers install AI coding assistants. Teams spin up automation agents. Business users connect AI tools directly to company data. Each of these is an AI agent running with some level of access to internal systems. Most organisations have no live inventory of what’s running, who deployed it, or what it can reach. Congress responded to this directly: the Stop Rogue AI Act, introduced September 3, 2026, would require organizations to maintain exactly that kind of machine-readable agent inventory.

The 18,000 vs 300 gap is extreme but not surprising. The same dynamic is happening in smaller businesses too. An AI agent that can read your CRM, write emails, and execute code is a meaningful attack surface. If it’s been installed by someone on the team without IT sign-off, that surface is invisible.

This is the governance gap that Falcon Guardian is designed to close. The key capabilities:

• Live agent inventory: a real-time map of what AI agents are running, who deployed them, what systems they access, and their security status
• Runtime monitoring: Falcon telemetry connects to agent activity so you can see what agents are actually doing, not just what they’re supposed to do
• Codex agent security: specific coverage for OpenAI Codex agents as they execute tasks across enterprise systems
• GPT-5.6 Cyber on Falcon: CrowdStrike’s platform gets access to OpenAI’s reasoning model trained for cybersecurity, improving threat detection speed and precision

What This Means for Business

If you’re running AI agents (and you almost certainly are, whether you know it or not) this announcement is a signal about where the market is heading.

Security vendors are no longer just protecting humans and their endpoints. They’re extending that protection to the AI systems acting on behalf of humans. The question of “what can my AI agent do, and is it doing it safely?” is becoming as important as “what can my employee do, and are they doing it safely?”

For businesses deploying AI agents through Omni Ops or similar platforms, the practical implication is this: agent security isn’t an afterthought you layer on after deployment. It’s something that needs to be designed in from the start, with clear scope boundaries, access controls, and monitoring in place before agents go live.

The 18,000 unauthorised agents story isn’t a failure of technology. It’s a failure of process. No organisation deliberately deployed 60 times more AI agents than they intended. It happened gradually, through individual decisions that seemed low-stakes at the time.

The businesses that get ahead of this will be the ones that build governance alongside deployment. Those that don’t will spend 2027 dealing with the kind of incident that ends up in the press release of a security vendor’s next conference.

---

This article is based on announcements made at Fal.Con 2026. Enterprise DNA covers AI news relevant to business leaders, data professionals, and teams deploying AI in production.

Building an AI agent workforce for your business? Enterprise DNA’s Omni Ops service deploys AI agents with governance and security built in from day one, not retrofitted after the fact.