The latest in AI, every dayAI News
← AI News

October 27, 2026 · Let's Data Science

Arrakis Security Report Finds 76% of AI Agents Dormant

My take: A report from cybersecurity company Arrakis Security found that 76% of active AI agents at the businesses they analyzed have not run in the last 60 days, yet their credentials and permissions still work exactly as they did on day one. 44% have gone unused for 180 days, and 82% of all agents reviewed carry at least one exposure of this kind.

This matters because many companies are already building AI agents to automate tasks (organizing files, moving data between systems, responding to requests) without thinking about who tracks that agent once it stops being used. When someone leaves a company, their access gets removed the same day. An AI agent nobody has used in months can still have active access to sensitive systems.

Agentic AI is a powerful tool, but like any new tool it needs the same basic rules of order we apply to any employee: know who created it, what it is for, and when it should be shut down. It is worth noting that this report was published by a company that sells security for AI agents, so its numbers are worth reading with that in mind, even though the underlying problem (abandoned agents with active access) is real and documented.

If your business already uses AI agents, do you have a list of which ones exist and who is responsible for shutting them down once they are no longer needed?

Read at the source: Let's Data Science ↗

Want to use these tools? See the unbiased reviews or back to the news.