Forgotten AI Agents Are Still Switched On and Still Holding the Keys

2026-10-27 • Technology

Forgotten AI Agents Are Still Switched On and Still Holding the Keys

Your office plant gets watered more reliably than your AI agents get retired. A new report found 76% of active agents had not run in 60 days, yet their credentials still worked and their triggers still fired. It is like leaving your spare keys with a stranger and then forgetting the stranger exists. Nobody removes them, says the Arrakis CEO, because nobody owns them.

76%Active agents idle for 60 days
44%Agents idle for 180 days
24%Dormant agents startable by outside webhook

The Ghosts in the Org Chart Still Have Badge Access

The 2026 Autonomous Agents Threat Report from Arrakis Security studied anonymized production data from thousands of AI agents over 12 weeks (CyberNewswire, 2026). Of the active agents, 76% had not run in 60 days, yet their credentials still worked and their triggers still fired. A total of 44% had not run in 180 days.

Think of a former intern whose badge still opens the server room. Nothing is happening, which is exactly why nobody checks. Arrakis CEO and co-founder Tal Baron said nobody removes anything, because in most companies nobody owns it (CyberNewswire, 2026).

What a Sleeping Agent Can Still Do

Dormant does not mean harmless. Among dormant agents, 24% could be started by an unsigned webhook from outside the organization, and 22% could send content to a third party hosted model and act on its response (CyberNewswire, 2026).

Another 20% could hand a task to an agent with different permissions and receive results they could not produce themselves (CyberNewswire, 2026).

Share of AI agents in percent, with the first two rows covering all active agents and the rest covering dormant agents
Idle for 60 days76%
Idle for 180 days44%
Startable by unsigned webhook24%
Can use outside model22%
Can borrow another agent20%

Who Owns the Robot Nobody Remembers?

A fair note of skepticism is in order. The report comes from a company that sells tools to scan and monitor agents, so the numbers deserve a raised eyebrow (CyberNewswire, 2026). The question it raises is still a good one, because unused agents with working credentials remain usable paths for an attacker.

The fix is dull but effective. Give every agent a named owner who can switch it off when it stops being useful.

Nothing is happening, which is exactly why nobody checks.

What You Can Actually Do About It

List the AI tools, automations, browser extensions and connected apps you have authorized, then remove the ones you no longer use and revoke their access. At work, ask your IT team who owns each agent and what it can reach. Prefer tools with limited permissions, and review them on a regular schedule. Then ask yourself, how many of the AI tools you connected to your accounts do you still use?

References

CyberNewswire. (2026, October 21). Arrakis Security report finds 76% of AI agents are dormant but keep their credentials and connections. Hackread. https://hackread.com/arrakis-security-report-finds-76-of-ai-agents-are-dormant-but-keep-their-credentials-and-connections/

How many of the AI tools you connected to your accounts do you still use?
Enjoyed this? Get the next one in your inbox

New articles on finance, technology, trade and real estate, written in plain English.

← Back to articles

Comments (0)

No comments yet — be the first to comment.


Leave a comment