Risks surface when forgotten AI agents run in the background

Governing the AI agents that organizations know about is difficult enough, but firms might face a hidden threat: agents they don’t know about. As employees begin using agentic AI more in their daily workflows, a company can easily lose track of the agents running in the background of their systems. And the problem can be compounded when an employee who launched an agent leaves the company. “Let's say Jane is in the HR department and she kicks off an AI agent inside the payroll application to help her with her job,” Frank Vukovits , chief security scientist at cybersecurity firm Delinea , told FinAi News . “What happens when Jane leaves the company? Does that agent still run in the background? Does that agent have the same access that Jane had?” Using AI agents securely requires strict human oversight, but a human cannot oversee an agent they do not know about. Secret agents To address this problem, identity and access management company Okta last month extended shadow AI discovery in its Okta for AI Agents product, allowing organizations to find unmanaged agents running on employee devices. [caption id="attachment_57389346" align="alignright" width="300"] (AI generated)[/caption] “You can't protect what you can't see,” Ric Smith , president of products and technology at Okta, said at a Sept. 17 news conference. “Securing unknown agents in the browser is crucial. Employees are spinning up local models and autonomous agents directly on their laptops and workstations.” Organizations should have strict policies in place for registering AI use to prevent the creation of secret agents, Vukovits said. By keeping track, companies can better limit an agent’s reach, only giving it the access it needs for its given task. An AI agent will likely have similar access and privileges to the employee who launched it — but probably doesn’t need to, Vukovits said. AI will keep going Regulating access and privileges at the outset is vital because AI agents typically do not stop without being told, Vukovits said. “A human is just going to do the three things they’re supposed to do and stop when the task is complete,” Vukovits said. “An AI agent, though, as we've seen in the news, may think a different way. It may not stop.” OpenAI confirmed this was the case in July when it announced that its advanced models had gone to “extreme lengths to achieve a rather narrow testing goal,” escaping their sandbox to infiltrate machine learning company Hugging Face . “The agent was given a task and then it figured out a way to get out to the internet,” Vukovits said. “It decided it needed to get out to the internet to complete the task, even though the sandbox restricted that.” To combat this, organizations should employ continuous authorization, rechecking an agent’s privileges and access at each step of a task to ensure it is not straying where it should not go, he said. FinAi Lending Summit, set for Oct. 7-8 in Las Vegas, will include speakers from Fifth Third and Capital One as well as a fireside chat with U.S. Bank Senior Vice President – Lending Operations Group Manager Suzanne Rathbun . To learn more about the 2026 event and register, visit here .

Risks surface when forgotten AI agents run in the background

Organizations must keep track of what agents employees are using

Bank of America uses ‘Orchestra’ tool to decide if task needs open-source or closed-source model

Related Posts

The Next Generation of Cyberthreats: Powered by AI

MVB Bank bolsters vendor risk management via Kobalt Labs AI

Banking’s strong safeguards, regulations lower exposure to verification-request fraud

EMERGING FINTECH DIRECTORY

FinAi Podcast

SPONSORED

The Next Generation of Cyberthreats: Powered by AI

Build an Antifragile Strategy to Outperform the Market

How AI and Product Experts Turn Fuzzy Requirements Into Focused Dev-ready Roadmaps

Connect

Welcome Back!

Retrieve your password

Continue Reading with FinAi News Premium - Less than $2/Day

Risks surface when forgotten AI agents run in the background

Organizations must keep track of what agents employees are using

Bank of America uses ‘Orchestra’ tool to decide if task needs open-source or closed-source model

Related Posts

FinAi Podcast

Connect

Welcome Back!

Retrieve your password

Unlock This Article

Continue Reading with FinAi News Premium - Less than $2/Day