An Osterman Research survey of roughly 380 knowledge workers asked one simple question: after you left the company, how many systems could you still get into? The answer was uncomfortable. 89% said they kept access to email, Salesforce, SharePoint, cloud storage and corporate accounts. Close to half admitted logging into one of those accounts after their last day, and 60% said nobody ever asked for their credentials on the way out.
The survey is old, and that is exactly the point: the failure is structural, not a passing trend. Back then, the list of forgotten systems held email and spreadsheets. Today it also holds AI assistants with access to internal documents, database integrations and entire conversations about company strategy.
This is not one department being careless. It is an architectural gap. When someone leaves on one side (the company directory) and removal on the other side (AI accounts, connected systems, cloud tools) depends on somebody remembering, you have an identity gap — and an identity gap is a security gap.
The size of the vacuum
In 2022, Oomnitza surveyed 213 senior IT professionals at companies with 1,000 to 10,000 employees and measured that gap with current numbers. 42% reported cases of unauthorized access to cloud applications stemming from incomplete deprovisioning — and another 17% simply could not say how large the problem was, which is worse than a bad number.
The same survey carried a second figure: 27% of companies lose more than 10% of their technology assets when an employee is offboarded — not because anyone steals anything, but because there is no record of what that person held, where it was, or what they were still connected to. And 48% of respondents pointed to missing or deficient automated workflows in offboarding. In other words: the process exists on paper, but it depends on people remembering.
Leaving that gap open now has a measured price. IBM's 2025 report on the cost of data breaches shows that 97% of organizations that suffered an AI-related security incident lacked proper access controls for those tools, and that 63% had no AI governance policy at all. In the same study, companies with a high level of AI used outside corporate management carried an extra USD 670,000 in average breach cost — that figure is about AI running outside the company's control, and an ex-employee account nobody switched off is precisely that.
And when the Verizon Data Breach Investigations Report analyzed more than 22,000 security incidents and 12,195 confirmed breaches in 2025, it found credential abuse as the way in for 22% of breaches. A credential that stays valid is a credential that keeps working for someone.
Why offboarding is different from termination
When someone leaves one side (the corporate directory, the payroll system), they need to leave all the other sides (email, cloud tools, AI assistants, internal systems). Termination is the event (the last day). Offboarding is the process.
The process has phases:
- Identity leaves the directory — removal from Active Directory, from the company's single sign-on, and so on.
- Access leaves the tools — revoke credentials in every connected system, AI ones included.
- Data leaves or gets archived — transfer or delete according to policy.
- Audit records everything — a trail of who accessed what while authorized.
When phase 1 does not connect to phase 2, you end up with someone off the payroll but still inside the AI. And phase 4 loses its meaning: a trail that does not know who should already be gone records the access, but cannot classify it as improper.
What begins to happen
An ex-employee has the password saved in her browser. She opens the company's AI assistant. She uses that access to pull reports from the finance team. Nobody notices, because the session comes from old local history and the system does not know she left.
Or the ex-employee had access to a critical database integration through a personal credential. The credential was never revoked. Months later, during an investigation, someone finds queries that match no active team. Where did they come from? There is no way to tell, because nothing in the system marks the moment that person stopped being authorized.
The expensive part is that second case: it is not only the improper login, it is the time lost figuring out what happened. The Ponemon Institute study on insider risk shows the size of the difference: incidents contained within 30 days cost the surveyed organizations USD 14.2 million a year, against USD 21.9 million when containment stretches past 90 days. The same study reports that 92% of organizations say generative AI has changed how people access and share data — and that only 18% have integrated AI governance into their insider risk program.
What changes when corporate identity governs everything
If the AI is plugged into the company's Active Directory — or any other corporate directory that is the source of truth for identity — termination and offboarding become the same thing. The instant someone leaves in the directory:
- AI access drops with them.
- It does not depend on anyone remembering to revoke an account, a password or a credential.
- It is on record who they were, when they lost access, and why.
- Any query or action after that moment cannot have come from that user — and that shortens any incident investigation.
It is the difference between "we hope people remember to shut down dozens of different tools" and "when identity leaves the corporate directory, every tool sees that this person no longer exists."
Skyller uses this mechanism: identity from the company's Active Directory, access according to each person's role, human approval before a sensitive action, and a record of who did what and when. Someone removed from the directory gets in nowhere.
What has to be in place
- If an employee is let go today, how many AI tools can they open tomorrow? If the answer depends on someone remembering to revoke accounts, the problem is identity, not policy. Measure how long access takes to drop — if the unit is "days", it is already too late.
- How many ex-employee accounts still exist in your AI tools? If nobody can answer, revocation is not automatic — and you are in the group of 17% who cannot measure their own exposure.
- Is there a personal credential per tool, or does the team share logins? A shared login survives any offboarding, because it belongs to nobody who can be offboarded.
- Does the audit trail know when each person stopped being authorized? Without that marker, the record shows the access but cannot establish that it was improper — which is exactly what an investigation has to prove.
