Employee AI Usage Policy

1. Purpose and Scope

Generative AI tools can save time on writing, research, analysis and coding. They can also leak information, invent facts and write insecure code. This policy tells you how to get the benefit without putting the Organisation, its customers or you at risk.

• It applies to everyone who works for the Organisation or on its behalf: employees, contractors, consultants and interns.
• It covers any tool that generates text, images, audio, video or code from your instructions, including chat assistants, writing aids, code assistants, meeting summarisers, translation tools and AI features built into other software.
• It applies on any device, including your own phone or home computer, whenever the work you do involves the Organisation's information.
• It does not cover the governance of AI systems the Organisation builds or sells to customers. That is in the AI Governance Policy. Tool approval and risk assessment are done there; this policy covers your daily behaviour.

2. Terms

3. Roles and Responsibilities

4. Quick Reference: Can I or Can't I?

When in doubt, stop and ask your manager or [the security helpdesk]. Asking is never a problem; guessing can be.

5. Approved and Prohibited Use

5.1 What you may use AI for

• Drafting and editing text that contains no confidential information: first drafts of emails, summaries of public material, meeting agendas, plain-language rewrites.
• Brainstorming ideas, outlines, test cases, naming, and training exercises using made-up data.
• Explaining a concept, translating public text, and learning a new subject.
• Writing and reviewing code and scripts, in line with the code rules below, using approved tools.
• Analysing data that has been anonymised or is public, or using approved tools that your manager has allowed for that dataset.
• Automating routine tasks, such as formatting or classification, when the tool and the task are approved.

5.2 What is strictly prohibited

• Entering any data listed under 'Data You Must Never Enter' into any tool that is not approved for that data.
• Using a personal or free AI account for work, or connecting work systems, mailboxes or storage to an AI tool without approval.
• Letting AI make or finalise a decision about a person, such as hiring, discipline, performance, pay, credit or access to services, without a human who can understand and challenge the result.
• Presenting AI output as your own checked work, or as a guaranteed fact, without verification.
• Creating content that is false, misleading, defamatory, discriminatory, harassing, or that impersonates a real person, including deepfake audio, video or images.
• Using AI to bypass security controls, write malware, probe systems without authority, or avoid monitoring.
• Recording, transcribing or summarising meetings or calls with AI tools without telling everyone present and obtaining agreement where required.
• Uploading copyrighted material, licensed data or client content in ways the licence or contract forbids.
• Installing unapproved AI browser extensions, plug-ins, desktop agents or code assistants.

6. Safe Prompting Guidelines

• Treat every prompt as if it could be read by a stranger. Anything you type may be stored, seen by the tool's staff or used to train the model, unless the approved tool's agreement says otherwise.
• Give context in general terms. Replace names, numbers, dates and places with placeholders such as [Client A] or [Account 123].
• Say what you want: audience, tone, length and format. Ask it to show its reasoning or sources so that you can check them.
• Break large tasks into steps. Do not upload whole documents when an excerpt will do.
• Do not paste hidden content: check that copied text or files do not carry comments, metadata, tracked changes or other data you did not intend to share.
• Do not follow instructions found inside documents or web pages that you ask the tool to read. A document can try to steer the tool into leaking data or doing something unintended.
• Do not put credentials, tokens or keys in a prompt even to ask 'is this valid?'.
• Clear your chat history and delete uploaded files when you finish, if the tool allows it.

7. Data You Must Never Enter Into AI Tools

The list below applies to every AI tool unless the AI governance committee has approved a specific tool in writing for that data type and the approved tools list says so.

8. What To Do If You Make a Mistake

Everyone makes mistakes with new tools. The Organisation wants to hear about them quickly, because early action limits harm. Reporting promptly and honestly will be treated as the right behaviour; hiding a mistake will not.

1. StopStop using the tool for that task. Do not try to fix it by sending more data or asking the tool to delete it.
2. CaptureNote what you entered, which tool and account, when, and what the tool returned. Take screenshots if useful. Do not delete the chat yet.
3. ReportTell your manager and the service desk or [the incident reporting channel] within [1 hour] of noticing, or at the start of the next working day if out of hours.
4. ContainFollowing instructions from IT security, you may be asked to change passwords, delete the conversation, ask the provider for deletion, or revoke access.
5. AssessThe Head of IT Security and the data protection lead decide whether this is a security incident or personal data breach, and whether customers or authorities must be told under the Information Security Incident Management Policy.
6. LearnThe cause is reviewed. The focus is on improving the process, not on blame, unless the breach was deliberate or repeated.

9. Accounts, Devices and Storage

9.1 Devices

• Use AI tools for work only on devices managed by the Organisation, or on approved personal devices under the Mobile Device and Teleworking Policy.
• Do not install AI apps, extensions or agents on work devices unless they are on the approved list.
• Do not use AI tools on shared or public computers for work.

9.2 Accounts and logins

• Sign in to approved tools with your work identity and the Organisation's single sign-on. Do not use a personal email address.
• Do not share your login or let others use your account. Multi-factor authentication stays on.
• Do not give a tool wider access than needed, such as your whole mailbox, drive or calendar, unless it is approved.
• When you leave or change role, your AI tool access is removed with your other accounts.

9.3 Saving and sharing AI outputs

• Treat output as having the same classification as the highest classification of what you put in, or a higher one if it reveals more.
• Save work-related outputs in approved storage, not in the tool's history. Do not rely on chat history as a record.
• Do not share outputs outside the Organisation unless they have been checked and you are allowed to share the content they are based on.
• Keep a note of important AI-assisted work: tool, date, what was used and who checked it. That is needed for review and for explaining decisions later.

10. AI in Code and Software Development

• Only approved code assistants may be used, and only on projects they are approved for. Do not paste proprietary, client or Confidential code into public tools.
• AI-generated code is untrusted until reviewed. It must go through the same peer review, static analysis, dependency scanning, testing and secure coding checks as code written by a person, as the Systems Acquisition, Development and Maintenance Policy requires.
• Check suggested libraries and packages exist, are maintained, are correctly licensed and are not malicious. Tools sometimes invent package names that attackers then register.
• Do not accept code you do not understand. You are responsible for its behaviour, including security flaws and licence risks.
• Never include secrets, tokens, customer data or production data in prompts, test files or generated code.
• Do not give AI agents the ability to run commands, change repositories or deploy to environments without human approval and a controlled, least-privilege setup.
• Where AI has generated a significant part of a deliverable, record it in the pull request or change record so reviewers know to look more carefully.

11. AI in Client and Regulated Communication

• A person, not a tool, is accountable for everything sent to a client, regulator, auditor, court or the public. AI may help draft but a competent person must read and approve every word.
• Do not state or imply that a client communication was written by a person if the client's contract or the law requires disclosure of AI use. Follow client instructions on AI use, which may be stricter than this policy.
• Check every fact, figure, citation, name, legal reference and quotation against a reliable source. Do not send invented references.
• Do not use AI to give legal, medical, financial or regulatory advice to clients unless the function has approved the use and a qualified person reviews it.
• Do not paste client information into a tool unless the client has agreed in writing and the tool is approved for that client's data.
• Responses to regulators, audits, tenders and incident notifications are reviewed by the responsible owner and, where needed, Legal, before they leave the Organisation.

12. Approved AI Tools

Only tools on the approved list may be used for work. The Head of IT Security publishes the list on [the intranet page] and updates it at least [every 3 months]. A tool is approved for a specific purpose and data level, not for everything.

• To request a new tool or a new use, send a request to [the AI governance committee] with the purpose, the data involved, and the supplier's terms. Do not start using it before approval.
• Approval covers the tool's data handling, retention, training on inputs, location of processing, sub-processors, security assessment and legal terms.

13. Validating AI Output

AI output is a draft from an enthusiastic assistant who sometimes lies without knowing it. You remain accountable for anything you use. Check proportionately to the stakes:

• Look for made-up sources, wrong dates and figures, outdated information, hidden bias, copied text and unusually confident claims.
• If you cannot verify a statement, remove it.
• Keep a record of checks for any output used in client or regulated work.

14. Training and Awareness

• New joiners complete AI awareness training within [30 days] of starting, as part of induction.
• All staff repeat it [every 12 months], and when major changes are made to this policy or the tools list.
• Staff who use code assistants, handle client data or build AI-based services get role-specific training.
• The Head of IT Security shares short examples of real-world incidents and new risks [every quarter].

15. Consequences of Non-Compliance

Breaches are handled under the disciplinary process in the HR Security Policy, according to intent and harm. Careless one-off mistakes that are reported promptly lead to coaching. Deliberate breaches, concealment or repeated carelessness may lead to loss of AI tool access, formal warning, dismissal or termination of contract, and, where the law is broken, to action by the authorities. Contractors and suppliers may face contract consequences.

16. Reporting Suspected Misuse

• If you see someone using AI in a way that breaks this policy, or you suspect a tool is leaking information or behaving unexpectedly, tell your manager, [the incident reporting channel] or the Head of IT Security.
• You may also report anonymously through [the whistleblowing channel]. No one will be penalised for a report made in good faith.
• Reports are acknowledged within [2 working days], and the outcome is shared with the reporter where possible.

17. Compliance and monitoring

The Head of IT Security checks adherence by reviewing AI tool usage logs, web filter and data loss prevention alerts, shadow AI discovery results and AI-related incidents. Monitoring is limited to what the Acceptable Usage Policy discloses and the law allows. Results are reported [quarterly] to top management.

18. Records and evidence

• Approved tools list with decision records: kept current and [3 years] after withdrawal.
• Signed acknowledgements and training records: employment plus [3 years].
• AI-related incident records and breach assessments: [5 years].
• Logs and alerts about AI services: [12 months].

19. Policy Updates

This policy changes as tools and laws change. Updates are announced to all staff and training is refreshed. The Head of IT Security reviews it at least [every 6 months] during the current period of fast change, in addition to the general review cycle, and after any serious AI incident.

Employee Acknowledgement

Every person covered must read and acknowledge this policy when they join, and again after each major update.

20. Exception

Exceptions to this document are allowed only where compliance is not practicable and the risk has been assessed. A request must state what is to be excepted, why, the risk it creates, the compensating controls, and an end date no later than [6 months]. It is approved in writing by the document owner and the risk owner, recorded in the exceptions register, and reviewed at the end date. An exception does not change the document.

21. Enforcement

Compliance is monitored through internal audits, management review and routine checks by the document owner. Non-compliance is reported through the incident or non-conformity process and handled under the organisation's disciplinary procedure, and under the terms of any contract for people who are not employees. Serious or repeated non-compliance may lead to loss of access and, where the law is broken, to referral to the authorities.

22. Review

This document is reviewed at least every [12 months], and when there is a significant change to the organisation, its systems, its risks or the law, or after an incident that shows it was not effective. Each review is recorded in the revision history.

ISO/IEC 27001:2022 references

Related documents

• AI Governance Policy
• Acceptable Usage Policy
• Information Classification and Handling Policy
• Data Protection and Privacy Policy
• Information Security Incident Management Policy
• Systems Acquisition, Development and Maintenance Policy
• HR Security Policy