October 31, 2026 · 2 min read
78% of companies took access back from an AI agent. Where is your off switch?
New research from Agen.co by Frontegg found that 78% of security leaders have pulled permissions from an AI agent already in production. A separate RealVNC survey found only 26.4% of organizations can revoke agent access. Here is how a small business can start narrow and keep a clear off switch.
If one of your AI agents started doing the wrong thing today, how quickly could you stop it? Would you know which button to press, and who is allowed to press it?
On October 27, Agen.co by Frontegg, a security software company, released a survey of 300 CISOs and CSOs at U.S. companies with 500 or more employees. The research firm Cint ran it in late September. 78% said their organizations had pulled permissions from an AI agent already running in production. Some did it after a security concern, others during a routine review. 49% had found an agent with access it should never have had, and only 37% limit an agent's access to the task it was built for.
A day later, RealVNC, a remote access software company, shared its own survey of IT and technical decision makers in the U.S., U.K., and Canada. Nearly three quarters of organizations let AI agents act without prior human approval, yet only 26.4% can revoke agent access.
Both studies come from software vendors and focus on larger organizations. Still, the lesson is a gentle one.
Why this matters for a small business
Pulling access back is not a failure. It is what good leaders do once they see how a new helper really works. The risk is giving an agent wide access on day one, with no easy way to take it back. In a small business, the person who set up the agent is often the only one who knows how to turn it off. That is a lot to hang on one person.
What I would do this week
- For each AI agent, write down what it can access, and remove anything it does not need for its one job.
- Write a short "how to pause this agent" note, and make sure two people on your team know where it is.
- Pick one agent and practice switching it off and back on, so you know it works before you ever need it.
This is Human-in-the-Loop AI. Start the agent small, widen its access as it earns your trust, and keep a person's hand near the switch. Automate the process, not the relationship.
My takeaway: Give every AI agent the smallest set of keys it needs, and make sure you can always take them back.
Sources
- 78% of Enterprises Have Pulled Permissions From an AI Agent Already in Production · Agen.co by Frontegg via PR Newswire (vendor research, October 27, 2026)
- Many Organizations Let AI Agents Act, But Few Can Stop Them · RealVNC via PRWeb (vendor research, October 28, 2026)