An employee leaves, IT disables the usual accounts and the organization considers offboarding complete. But an AI assistant might still work—and company information left in its conversations could be at risk and harder to recover.
A survey published in October by managed IT and cybersecurity provider Sagiss found that 24% of former employees reported they could still use a company AI tool or assistant after leaving. Another 16% said company information remained in an AI tool or chat history they had used for work.
The survey was conducted in September through Pollfish and included 1,000 U.S. adults who had left jobs involving company technology within the previous five years.
These are self-reported experiences, not verified audits of corporate accounts. But they raise a practical question for CIOs: How does an organization know that access has actually ended?
The account can close while the data remains
The survey didn't establish what kinds of AI accounts former employees could still use, according to Travis Springer, president of Sagiss.
Some might have been company-managed AI accounts that weren't disabled. Others might have involved personal AI accounts used for work. Springer said the survey couldn't determine which situation was more common, but the distinction matters because the two problems require different responses.
"Disabling an account is only part of the job. You also have to think about what company information has already been put into these tools and who controls it."
Travis Springer
President, Sagiss
A company can disable an account it manages, but information already copied into a personal AI tool might remain available through an account outside its control.
"Disabling an account is only part of the job. You also have to think about what company information has already been put into these tools and who controls it," Springer said.
The challenge extends beyond standalone AI assistants.
Security consultant Dave Shackleford has identified additional security risks in AI-enabled SaaS applications, which can query business systems, modify records and invoke APIs on a user's behalf.
Shackleford also pointed to service accounts, API tokens and OAuth grants as potential security blind spots. Some can be long-lived or poorly tracked. AI agents can inherit employee permissions, expanding the access available through those connections.
Organizations therefore need to inventory embedded AI capabilities, connected agents and nonhuman identities—not just the applications employees use.
Source: Sagiss, 2026 Managed Security Report: Employee Offboarding. Survey of 1,000 U.S. former employees conducted through Pollfish in September 2026. Findings are self-reported.
IT can't close accounts it doesn't know about
The problems aren't necessarily technical.
Springer said Sagiss sometimes discovers accounts associated with departed employees during routine reviews because IT was never informed that those employees had left.
"We regularly review clients' inactive accounts and identify people who left the company without IT ever being notified," Springer said.
"HR knows the employee is leaving, the manager knows which tools that person used, and IT manages the accounts it knows about. Sometimes those pieces never come together."
Travis Springer
President, Sagiss
The survey found that 33% of respondents said responsibility for terminating their access was unclear or that nobody appeared to be handling it.
Springer described an organizational handoff in which different departments possess different information.
"HR knows the employee is leaving, the manager knows which tools that person used, and IT manages the accounts it knows about. Sometimes those pieces never come together," he said.
Some CIOs are already tackling the broader challenge of identifying AI tools employees use, including those introduced without formal IT approval.
At Flexera, CIO and CISO Conal Gallagher said the company surveyed employees to identify informal AI use and determine which tools were helping them work. The effort surfaced applications that business teams had adopted independently.
"Our first step was to bring informal AI use cases into the open by engaging employees," Gallagher told TechTarget, describing Flexera's approach to discovering shadow AI.
At cryptocurrency payments company MoonPay, CIO and CISO Doug Innocenti described a more technical approach. The company used AI security software to examine applications and assess their underlying models, including how providers might use company data.
"We were able to perform an independent review on tools and understand which ones we wanted to pick," Innocenti explained to TechTarget’s Beth Pariseau.
The approaches differ, but both address a prerequisite for reliable offboarding: knowing which applications employees are using and where company information might flow.
Springer thinks HR should see the offboarding process through, even if IT and security are the ones shutting down accounts and checking that access is gone. Companies can divide up those responsibilities differently. The trouble comes when no one follows up on what's been missed.
Among respondents to the Sagiss survey who had used seven or more work accounts, 87% said at least one still worked after they left. Even among those who had used just one to three accounts, 71% reported the same problem.
Disabling an account isn't always enough
Single sign-on (SSO) makes offboarding easier. IT can turn off one central account instead of tracking down logins for every connected application. The question is what happens to access that's already been granted.
Microsoft notes that shutting off an employee's access doesn't necessarily end every active session right away. Some applications may still accept previously issued access tokens or maintain their own sessions. That means IT can't assume disabling a central account immediately cuts off access everywhere.
Springer recommends checking applications outside SSO, shared credentials and sessions that might remain active. He also wants CIOs to consider what company information employees have already stored in AI services and whether administrators retain control over it.
That requires more than a completed offboarding ticket.
Springer said organizations should document which accounts were disabled, when they were shut down and who verified the changes. They should also investigate delays, missed applications and unexpected login activity after an employee leaves.
The larger challenge is determining where an organization can enforce access removal and where its control over previously shared information is limited.
"You should be able to confirm someone is off the systems instead of finding out weeks later that an account still works," Springer said.
James Alan Miller is a veteran technology editor and writer and lead editor for CIO news at TechTarget. He directs coverage of enterprise technology strategy, AI, software, data, infrastructure and the decisions shaping how CIOs manage increasingly complex IT environments.
