How do I remove a former employee's access to AI tools and connected business data?

25 October 2026

How do I remove a former employee's access to AI tools and connected business data?

Treat offboarding as an access map, not a single account deletion. At cut-off, disable core identity and sessions; then remove AI workspace membership, direct system accounts, integrations, tokens and shared secrets. Preserve necessary business records under company control, test transferred workflows and keep evidence that each high-risk route was closed.

What should happen first?

Start with the identity account that grants access to other systems. At the agreed leaving time, disable the employee's Microsoft 365 or Google Workspace sign-in, revoke active sessions, remove recovery methods and reset any credentials they knew. If departure is contentious, coordinate the exact time with HR and the business owner. Do not announce an access cut-off before the responsible person is ready to execute it. For an ordinary planned departure, prepare the checklist in advance and make one named person responsible for confirming each step.

Disabling one account is not the whole job. A ChatGPT, Claude, Gemini or Copilot account may have been created with a personal email, a separate password or a direct invitation. The employee may also hold an API key, a browser extension, an automation token or a shared password. Write down every route into business data before you declare the person offboarded. The NCSC's identity guidance explicitly calls for a joiners, leavers and movers process and says access should be revoked when no longer needed. Read the NCSC guidance.

As a practical target, disable sign-in and sessions at the leaving time, then work through connected systems the same day. This is an operational target, not a statutory UK deadline. Keep a timestamped record of what was disabled and who checked it. The 2025/26 UK Cyber Security Breaches Survey found that 43% of businesses identified a breach or attack in the prior 12 months. That figure is not an offboarding failure rate, but it is a reason to treat unused access as a real risk rather than an administrative nicety. See the government survey.

Where can AI access survive after the main login is disabled?

Check four layers: individual AI accounts, shared workspaces, connected applications and machine credentials. In the AI service, remove the employee from the organisation or team, revoke their sessions and check whether they are an owner or billing administrator. In your identity provider, remove app assignments and any single sign-on access. In Microsoft 365 or Google Workspace, inspect app consents, add-ons and OAuth connections; a third-party tool may have been granted access to email or files. In the CRM, accounts package, helpdesk and project system, remove the user's direct account and any delegated permissions.

Next inspect automations. A workflow in Zapier, Make, Power Automate or a custom agent may be running under that person's account. Simply deleting it can break customer follow-up or invoicing. Transfer ownership to a business-controlled service identity, test the workflow, then revoke the old connection. If a vendor does not support clean ownership transfer, document the current integration, create a new one under company control and retire the old one after testing. Never leave an ex-employee's personal login as the permanent foundation of a business process.

Finally rotate secrets that the employee could have copied: API keys, webhooks, shared passwords, recovery codes and access tokens. Rotate in a controlled order so live services do not fail. First create a replacement credential, update the integration and test it. Then revoke the old credential. For high-risk departures, prioritise credentials with write access to customer records, email or payments. For low-risk departures, the same inventory still applies; a pleasant exit does not change the technical fact that access persists until removed.

What about chat history, files and ownership?

Separate business records from personal information. Identify shared assistants, prompt libraries, knowledge bases, uploaded documents and conversation histories that staff need to continue work. Transfer ownership where the service allows it, export required business material through approved administrator features and set an authorised owner. Avoid logging into a former worker's personal AI account. If essential records sit there, work with HR and your data protection lead on the lawful, proportionate route to recover them.

Do not automatically erase every record associated with the leaver. The ICO says organisations should apply retention schedules and review personal information on departure; some employment records may need to be retained for references, pension or legal obligations. Keep only what there is a legitimate reason to retain, restrict access and dispose of it on schedule. Equally, do not retain all AI conversations forever on the vague theory that they might be useful. A conversation containing customer data can be both a business record and a privacy risk. See the ICO audit framework.

Set a clear handover owner for each active AI workflow. Record what the tool does, which data it reads, who approves its output and who receives failure alerts. This is where small firms often discover that a supposedly shared automation only worked because one employee paid for it on a personal card. Budget for the legitimate replacement account rather than preserving an insecure workaround. The cost may be one additional licensed seat or a modest monthly subscription, but the exact price depends on the tool and contract. Do not invent a licence cost before checking the supplier.

How do you check that access is really gone?

Use two passes. First, the administrator confirms the user is disabled or removed in each service and that active sessions have been revoked. Second, an independent owner verifies the result from the business side: can the former employee still appear in the member list, own a bot, approve a workflow or receive a recovery email? Inspect recent sign-in logs and API usage for the old identity and keys. Where logs are unavailable, record that gap rather than assuming absence of activity proves absence of access.

For a ten-person business, a simple spreadsheet is enough. Use columns for service, account identifier, role, connected data, owner, removal action, time, evidence and checker. Include systems that are easy to forget: meeting transcription bots, browser extensions, shared cloud folders, code repositories, social scheduling, payment tools and supplier portals. One person executes; another reviews high-impact items. The ICO recommends reviewing users' access rights when an employee changes role or leaves and keeping records showing that the adjustment happened. The ICO explains the expectation.

If you find access was still active after departure, stop it immediately, preserve relevant logs and assess whether there was unauthorised access to personal data. A live credential alone is a weakness; it does not automatically mean a reportable breach. If personal data was accessed or disclosed, follow your incident process and assess notification duties promptly. Do not quietly delete evidence to make the checklist look complete. Use the finding to improve the joiners, movers and leavers process.

When is a full technical offboarding exercise not needed?

If the person never had an AI account, never connected an app and never held business data, do not manufacture a complex AI-specific project. Confirm those facts against your account and integration inventory, remove their normal business access and record the result. Similarly, do not buy an expensive identity management platform merely to solve a ten-person checklist. A documented process, business-owned accounts and periodic reviews can be proportionate. If you have hundreds of accounts or many automated integrations, central identity management and automated deprovisioning become more valuable.

Do not confuse 'not needed' with 'we do not know'. If staff have been free to use personal ChatGPT accounts, browser extensions and company cards without approval, the inventory is incomplete. Start by asking which tools were used for work and checking expense, SSO, browser-management and app-consent records. Treat that discovery respectfully. A blame-heavy process discourages people from telling you where the important data and workflows actually sit.

The counterargument is that immediate deletion is safest. It can be, for access, but deleting data or integrations without a handover can destroy records and interrupt customers. The better sequence is to cut the person's access immediately, preserve legitimate business continuity under a controlled account, and then apply your retention and deletion rules. If an employee leaves unexpectedly and you cannot establish what was connected, isolate the highest-risk credentials first and bring in your IT provider or security specialist for a targeted review.

What is a workable small-business checklist?

Before the leaving date, name an owner and list every service where the employee has an account, including AI workspaces and connected data systems. Identify admin roles, shared passwords, API keys, automations and company-paid personal subscriptions. Record the agreed cut-off time. Prepare replacements for business-critical workflows so the cut-off does not accidentally stop incoming enquiries or customer updates.

At cut-off, disable core identity and remote access, revoke sessions, remove AI workspace membership and direct system accounts, and reset shared secrets. Move or replace business-owned integrations. Then inspect recovery addresses, app consents, webhooks, service accounts and scheduled automations. Confirm with a second person that the old identity no longer appears in critical access lists. Save a short evidence record rather than a screenshot of sensitive customer information.

Within the next working day, review sign-in and API logs, test transferred workflows and check ownership of knowledge bases, files and billing. Within a week, close remaining low-risk accounts and update the tool inventory. Repeat access reviews routinely, not only when somebody leaves. If you want help building a proportionate checklist for your existing tools, an initial review should produce an account map, responsible owners and a tested cut-off procedure, not just a generic policy document.

Is This Right For You?

This checklist is useful if staff have used AI workspaces, connected apps or automations with company information. A small firm can manage it with an inventory and two-person review; a larger estate may need central identity tooling. If no AI access existed, confirm that through your records and focus on ordinary offboarding. Do not purchase a platform to replace a basic process you can already run reliably.

Frequently Asked Questions

Is disabling Microsoft 365 or Google Workspace enough?

No. It may remove single sign-on access, but direct accounts, active sessions, API keys and shared passwords may still work. Check each connected service.

Should I delete the former employee's AI conversations?

Not automatically. Retain business records only where justified, restrict access and follow your retention schedule. Assess personal data separately.

What if an automation runs under their personal account?

Move it to a business-controlled account, test the replacement, then revoke the old connection. Record the new owner and failure alerts.

Do I need to rotate every API key?

Rotate keys the person could access, prioritising those with write access or sensitive data. Replace and test integrations before revoking a key where interruption would harm operations.

How quickly must access be removed in the UK?

There is no single universal statutory number of hours for every account. Set the cut-off at departure, act promptly and document your controls under your security and data protection obligations.

What if I discover access remained active after departure?

Disable it, preserve logs, check for use and assess whether personal data was accessed. Escalate through your incident process if needed.

Can I log into their personal AI account to retrieve business work?

Do not assume you can. Ask HR and your data protection lead to establish a lawful, proportionate recovery route and avoid accessing unrelated personal content.