AI agents made by staff frequently use the access of their creator
Nudge Security says the median company with agents has 7, and logs show the employee, not the agent.
Nudge Security, a security company, published its AI Adoption in Practice 2026 report. It uses anonymous data from its customers, and it tracks which AI tools they use and what the tools connect to. The company says agents that staff make in project trackers and customer record tools frequently operate with the credentials of their creator. Thus the agents have the same access to company data.
How it works
The agent frequently uses the OAuth grant or the access code of the employee. OAuth is a method to let an app work for a user. The grants stay until a person removes them. Endpoint tools do not find these agents because they operate in the platforms.
The numbers
In organizations with agents, the median had 7, and about 44% had more than 10. The agents operated on 15 platforms. Pasted content or files were in 18% of the observed prompts.
What is not known
The data come only from Nudge Security customers. The report gives no count of injury from these agents. It says most data that enters AI tools is a side effect of usual work.
Insight: Agents work as a person, and logs cannot name them
Cequence Security found that many agents that work for customers do not say that they are agents. Nudge Security finds that logs show the employee and not the agent. The two reports seem to show one problem. Agents work with a human identity, and logs cannot tell who did the work. Give each agent a separate identity before you add more.
Sources
Posted