Before using AI on a work task, find the rule that covers the exact tool, account, purpose and data. A colleague’s habit or a product’s enterprise branding does not answer those four questions.
This is a document-finding task. You are trying to leave with a usable answer, a source for it and a named person to ask about any gap.
Find the policy and its owner
Search the intranet and employee handbook for “AI,” “generative AI,” the product name, “acceptable use” and “approved software.” Check whether a separate tool catalogue names the permitted account or company tenant. Onboarding instructions can point you toward the policy, but check that they still refer to the current version.
Record the document title, owner, effective date and link in an approved work location. If versions conflict, ask the owner which governs your task. Do not copy a confidential policy into a personal chat to have it interpreted.
The ICO’s published internal AI policy is a concrete example: it separates approved tools and devices from permission to process personal, sensitive or confidential information. It also identifies review and incident routes. It governs the ICO, not your employer; use it to understand the kinds of questions a policy can answer.
Read for scope, not a general yes
| Question | What a usable answer names |
|---|---|
| Which tool? | Product and approved account or tenant |
| For what work? | Task, purpose and output audience |
| With what input? | Permitted and prohibited data categories |
| With which connections? | Uploads, integrations and recipients in scope |
| Before use or release? | Approval, verification, disclosure and record requirements |
| If something goes wrong? | Named incident route and first action |
Ireland’s 2026 public-sector AI deployment guidance likewise includes acceptable-use rules, training and incident reporting. Those public-sector controls are examples, not a universal rulebook for every workplace. NIST’s risk-management framework provides the broader organizational context if you own the policy rather than merely need to find it.
A completed scope check
This is a fictional teaching example, not an interpretation of a real employer’s policy.
A tool catalogue permits “Company Assistant, company login, drafting from public information.” An employee wants to summarize client interview notes.
- Tool and account: match the catalogue.
- Task: summarization is not explicitly covered by the stated drafting permission.
- Data: private client notes are not public information.
- Result: the catalogue does not establish permission for this use.
The next step is a focused question to the authorized owner, without attaching the interview notes:
“Does Company Assistant cover summarizing confidential client interviews for an internal project report? The catalogue entry only names drafting from public information. If this use is permitted, which data, review and retention conditions apply?”
A different task, drafting from an already public press release, may fit that rule. Do not turn uncertainty about one use into either blanket permission or a blanket ban on every use.
When the policy is missing or incomplete
Existing confidentiality, security, procurement, privacy and client rules still apply. Ask the person authorized to resolve the gap, which may be IT, security, privacy, procurement or a delegated manager. Keep sensitive material out of the proposed tool while its authorization is unresolved. Save the written answer with the policy reference.
Provider controls answer a separate question. ChatGPT data controls describe training settings; they cannot grant your employer’s approval. Likewise, the ICO’s data-minimisation guidance explains limiting personal data to the purpose, but does not decide the lawful basis or contract terms for your project.
Use the policy read card to document the answer. If an accidental paste has already happened, follow the workplace incident route promptly rather than trying to settle it through another chat.



