[AI Burst] Your AI Agents Have More Access Than Your Staff. And No One Is Watching.

AI Insights in 4 Minutes from Global AI Thought Leader Mark Lynd

Welcome to another edition of AI Bursts. One idea, analyzed fast, with what it means for you.

A single, powerful AI idea, analyzed rapidly.

AI agents are no longer experiments. They run real workflows, and they do it with broad access. The latest Darktrace State of AI Cybersecurity 2026 report puts a number on the worry. 92% of security leaders are concerned about AI agents across their workforce.

The reason is access. One agent often holds permissions across many systems at once. Sensitive data, business apps, tokens, APIs, even the security tools themselves. That is more reach than most employees get on their first day. And the agent has no manager, no badge, and usually no log.

Here is the gap. 92% are worried, but only 37% of organizations have a formal AI policy. That number fell from last year. Adoption did the opposite. Gartner expects more than 80% of enterprises to run generative AI in production by the end of 2026, up from less than 5% in 2023. McKinsey already finds 78% using it in at least one business function.

So access is spreading fast while governance falls behind. Security leaders know where it hurts. 61% name exposure of sensitive data as their top concern. 56% point to policy violations. 51% worry about misuse of the tools. An agent with broad access and no monitoring is the cleanest path to all three.

Treat every agent as an identity, not a feature. Three moves this week.

First, inventory the agents already running in your environment. You will find more than you expect. Second, cut their access to least privilege. An agent should reach only the systems its job needs, nothing more. Third, log and monitor what they do. If you cannot see an agent's actions, you cannot defend against them.

None of this is new security thinking. It is the same identity discipline you already use for people. Agents just never got added to the list.

The high-leverage stack you need to know this week.

The Identity Layer. Okta and Microsoft Entra. Both now extend identity governance to non-human and agent identities. You can assign, scope, and revoke an agent's access the same way you do for staff.

The Behavior Monitor. Darktrace. Its models watch for activity that drifts from normal. That is how you catch an agent moving data sideways or probing systems it should not touch.

The Shadow-AI Control. Netskope and Zscaler. Their security service edge platforms can see and block unapproved AI tools at the edge, so sensitive data does not leave through an app no one sanctioned.

Your 30-second scan of the AI landscape.

Adoption. Gartner expects more than 80% of enterprises to have GenAI in production by the end of 2026, up from under 5% in 2023.

Usage. McKinsey reports 78% of organizations now use GenAI in at least one business function.

Governance gap. Only 37% of organizations have a formal AI policy, a slight drop from last year, even as agent use climbs.

The idea of least privilege is not new. Jerome Saltzer and Michael Schroeder defined it in their 1975 paper on protecting information in computer systems. Give each part of a system only the access it needs to do its job. Fifty years later, we are still relearning it. This time the part of the system is an AI agent.

The Lesson. The fix for new technology is often an old principle applied honestly.

"These agents must be governed as identities, with least-privilege access and ongoing monitoring."

Darktrace, State of AI Cybersecurity 2026

Till next time,

Mark

The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.

Try Wispr Flow free