Generative AI Policies for Employees
Generative AI policies let employees use AI productively while protecting confidential information, personal information and privilege. A sound acceptable-use policy lists approved tools, sets information rules, requires human verification, addresses IP in outputs, covers disclosure duties such as Ontario's AI job posting rule, and is supported by training.

Key takeaways
- Approved enterprise tools with contractual protections are the key control.
- Information rules should follow the organization's classification scheme.
- Privileged material in public AI tools risks a loss of privilege.
- Employees remain responsible for verifying AI outputs.
- Ontario requires disclosure of AI use in publicly advertised job postings from 1 January 2026.
- Policies need training, monitoring and at least annual review.
Why generative AI policies matter
Employees are already using generative AI to draft emails, summarize documents, write code, analyze data and prepare presentations, often with tools the organization has never reviewed. Without clear rules, confidential information, personal information and privileged advice can end up in third-party systems, outputs can be inaccurate or infringing, and the organization cannot explain how important work was produced. Well-drafted generative AI policies give employees permission to use these tools productively while setting boundaries that protect the business, its clients and the people whose information it holds.

This insight walks through the core elements of an acceptable-use policy for Canadian employers: approved tools, confidential and personal information, privilege, intellectual property in outputs, disclosure of AI use, and training. It is general information, not legal advice.
Scope and definitions
A policy should define what it covers. Most generative AI policies apply to any tool that generates text, images, audio, video or code in response to prompts, whether stand-alone services, features built into productivity suites and other enterprise software, browser extensions or mobile apps. The policy should apply to employees, contractors and anyone else using the organization's systems or data, and to work done on personal devices. It should also explain its relationship to existing policies on acceptable use of technology, information security, privacy, records management, social media and the code of conduct, so that employees understand that the existing rules continue to apply.
Approved tools and the procurement gate
The single most effective control is distinguishing between approved enterprise tools and everything else. Enterprise versions of generative AI services often offer contractual commitments that inputs will not be used to train the provider's models, data retention controls, administrative oversight, single sign-on and audit logs, sometimes with Canadian data residency. Consumer versions frequently do not. The policy should list approved tools, or point to an internal register, and state the categories of information that may be used with each. New tools should go through a short review by IT, security, privacy and legal before use, with a clear route for employees to request them. A blanket ban rarely works: employees tend to use unapproved tools anyway, which is why a sanctioned alternative matters.
Rules for confidential and personal information
Generative AI policies should be explicit about what may not be entered into AI tools, or may be entered only into specific approved tools. A practical way to express this is a tiered table tied to the organization's information classification scheme:
| Information type | Public or consumer tools | Approved enterprise tools |
|---|---|---|
| Public information | Permitted | Permitted |
| Internal business information | Not permitted | Permitted |
| Confidential client or third-party information | Not permitted | Permitted only if contracts allow and the tool is approved for it |
| Personal information | Not permitted | Permitted only for approved purposes with privacy review |
| Sensitive personal information (health, financial, biometric, children's) | Not permitted | Only with specific approval |
| Privileged legal advice | Not permitted | Only in tools approved by legal |
Personal information rules matter because privacy law continues to apply. The principles for generative AI published by the Office of the Privacy Commissioner of Canada and its provincial and territorial counterparts call on organizations using generative AI to have legal authority for collecting and using personal information, limit use to appropriate purposes, be open about their use of AI, ensure accuracy and safeguard information. Entering customer or employee information into a tool that retains it for its own purposes may be an unauthorized disclosure.
Privilege loss from public tools
Solicitor-client privilege protects confidential communications between a client and its lawyer for the purpose of legal advice. Confidentiality is essential to privilege, and disclosing privileged communications to a third party can put privilege at risk. Pasting a legal memo or advice into a public AI tool whose provider may store, review or reuse inputs creates an argument that confidentiality, and therefore privilege, was lost. Even where privilege survives, the organization may face a dispute over it. The policy should prohibit entering privileged material into any tool not approved by the legal department, and legal teams should assess enterprise tools specifically for privilege-sensitive work. Litigation hold obligations should also extend to AI prompts and outputs where they are relevant.
Accuracy and human review
Generative AI can produce confident but false statements, invented citations, outdated information and flawed code. Courts and tribunals have criticized filings that relied on fictitious AI-generated authorities, and several Canadian courts have issued practice directions on AI use in court submissions. The policy should make clear that the employee who uses AI remains responsible for the output, must verify facts, figures and sources, and must apply professional judgment before relying on or sending any AI-assisted work. Higher-risk uses, such as advice to clients, regulatory filings, financial reporting or decisions about individuals, should require documented human review.
Intellectual property in inputs and outputs
Two intellectual property issues arise. The first concerns inputs: employees should not upload third-party copyrighted content, confidential information received under a non-disclosure agreement or licensed data unless the licence permits it. The second concerns outputs. Canadian copyright law protects original works of authorship, and the Supreme Court of Canada has held that originality requires an exercise of skill and judgment by an author. Whether, and to what extent, AI-generated material is protected by copyright, and who owns it, is unsettled in Canada. Outputs may also reproduce protected material from training data.
The policy should state that work product created by employees using AI tools in the course of employment belongs to the organization to the extent any rights exist, require employees to follow the tool's terms on output use, and restrict the use of AI-generated content in products, publications and code where IP ownership or infringement risk matters. For software development, organizations should apply open source and code provenance controls to AI-generated code. See our copyright and trade secrets practice.
If you deliver work to clients, check your client contracts. Some now restrict the use of AI tools on client matters, require disclosure, or prohibit entering client data into third-party AI services. The policy should require employees to follow client-specific restrictions.
Disclosure of AI use
When should employees disclose that AI was used? Generative AI policies should set expectations internally (for example, labelling AI-assisted drafts) and externally. External disclosure may be required by contract, by professional rules, by court practice directions, or by law.
In Ontario, since 1 January 2026, section 8.4 of the Employment Standards Act, 2000 requires an employer that advertises a publicly advertised job posting and uses artificial intelligence to screen, assess or select applicants to include a statement disclosing that use in the posting, subject to prescribed exceptions. Under Quebec's private-sector privacy law, organizations must inform individuals when a decision about them is based exclusively on automated processing of their personal information. Customer-facing chatbots should be clearly identified as AI, consistent with the federal voluntary code of conduct for generative AI.
HR and workplace uses
Using generative AI in hiring, performance management, discipline or workplace investigations raises particular risks under human rights, employment and privacy law. Outputs may reflect bias, and decisions about individuals require fairness and explainability. Generative AI policies should require HR uses to be approved, assessed for bias and privacy risk, and subject to human decision-making. Employers should also consider whether monitoring of employees' AI use is consistent with workplace privacy laws and any electronic monitoring policy requirements, such as Ontario's requirement for certain employers to have a written policy on electronic monitoring of employees. See our employment law practice.
Customer-facing and marketing uses
Using generative AI to create marketing copy, images, product descriptions or customer communications raises consumer protection and advertising issues as well as IP questions. Claims generated by AI must be accurate and substantiated, just as if a person had written them, and the Competition Act's prohibitions on false or misleading representations apply regardless of how content was created. Synthetic images of people or testimonials can mislead consumers if not handled carefully. Customer-facing chatbots raise questions of liability for incorrect statements. The policy should require marketing and customer-facing AI uses to go through the same review as other public content, with additional checks for accuracy and rights clearance. See our insight on AI chatbots and liability.
Records, retention and prompts
Prompts and outputs can be business records. They may contain personal information subject to access requests, information relevant to litigation, or evidence of how decisions were made. The policy should state whether prompts and outputs are retained in approved tools, for how long, and how they fit within the records retention schedule. Employees should not use personal accounts for work-related AI use, because the organization then cannot preserve, search or delete those records when required.
Security considerations
Generative AI creates new security risks: prompt injection through documents or web pages, malicious browser extensions posing as AI tools, data leakage through plug-ins and connectors, and AI-generated phishing. The policy should require the use of approved integrations only, prohibit connecting AI tools to corporate systems or data sources without review, and direct employees to report suspected AI-related security incidents through the usual channels.
Training and awareness
A policy that employees do not understand will not be followed. Training should explain the reasons behind the rules, show practical examples of acceptable and unacceptable prompts, describe approved tools and how to request new ones, and cover verification of outputs. Role-based modules for legal, HR, marketing, software development and client-facing teams make the training relevant. Short refreshers as tools and rules change keep it current.
Monitoring, enforcement and review
The policy should explain how compliance is monitored, consistent with privacy law and any notice requirements, and the consequences of breaches, aligned with the existing disciplinary framework. Because the technology changes quickly, assign an owner and review the policy at least annually, or sooner when new tools are adopted or the law changes. Collecting feedback from employees helps identify where the rules are unclear or where new approved tools are needed.
Checklist for generative AI policies
- Define scope and link to existing policies.
- List approved tools and the review process for new ones.
- Set information rules tied to classification levels.
- Prohibit privileged material in unapproved tools.
- Require human verification and accountability for outputs.
- Address IP in inputs and outputs and client restrictions.
- Set disclosure rules, including Ontario job posting and Quebec automated decision requirements.
- Cover HR uses, security, training, monitoring and review.
How TRW & Co helps
We draft and update generative AI policies for Canadian employers, including acceptable-use rules, tool approval processes, privacy and privilege safeguards, IP provisions, HR use rules and training materials, and integrate them with broader AI governance. See our artificial intelligence practice or contact us. Related reading: our insight on AI governance programs.
Frequently asked questions
Should we ban generative AI at work?
Outright bans are hard to enforce and often push use to unapproved tools. Most organizations approve enterprise tools with appropriate contractual protections and set clear rules for what information may be used.
Can employees put personal information into AI tools?
Only into approved tools for approved purposes after privacy review. Privacy law continues to apply, and entering personal information into tools that retain it for their own purposes may be an unauthorized disclosure.
Who owns AI-generated work?
The policy should provide that work product created in the course of employment belongs to the organization to the extent rights exist. Whether AI-generated material attracts copyright in Canada is unsettled.
Must we disclose AI use in hiring?
In Ontario, since 1 January 2026, employers that use AI to screen, assess or select applicants for publicly advertised job postings must disclose it in the posting, subject to prescribed exceptions.
How often should the policy be updated?
At least annually, and whenever significant new tools are adopted or the legal landscape changes.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


