How to Write a Generative AI Acceptable Use Policy for a Canadian Business in 2026

A generative AIgenerative AI — Generative Artificial IntelligenceAI that produces new content — text, code, images, structured data — rather than only classifying or predicting from existing inputs. acceptable use policy is the document that tells staff which tools they may use, what they may never put into them, and who checks the output. Writing it from a blank page invites gaps. Canada already has public, free guidance that maps closely onto the sections such a policy needs, so the faster route is to build each section from a named source and keep that source beside it.

This outline draws mainly on these official sources: the Government of Canada's Guide on the use of generative artificial intelligence, the federal, provincial and territorial privacy regulators' Principles for responsible, trustworthy and privacy-protective generative AI technologies, the Office of the Privacy Commissioner of Canada's (OPC) Guidance on assessing third-party service providers, Quebec's private-sector privacy Act and the Commission d'accès à l'information du Québec (CAI) pages that explain it, and the US National Institute of Standards and Technology's (NIST) Generative AI Profile.

This is not legal advice. It is a drafting outline that points to public guidance. Your obligations depend on your sector, your provinces of operation and the data you handle, so involve legal counsel and your privacy officer before you adopt a policy.

One point on scope: the Government of Canada guide is written for federal institutions and public servants, not private businesses. A company adopts its practices voluntarily. The privacy principles, the OPC service-provider guidance and the Quebec Act speak directly to private organizations.

1. Purpose and scope. State why the policy exists, who it covers and which tools count as generative AI. NIST AI 600-1 is useful here because it expects acceptable use policies to reach beyond employees: one of its suggested actions is to "update GAI acceptable use policies to address proprietary and open-source GAI technologies and data, and contractors, consultants, and other third-party personnel." Name contractors and consultants in your scope clause. The Government of Canada's FASTER principles (Fair, Accountable, Secure, Transparent, Educated, Relevant) also make a compact statement of intent; the "Relevant" principle asks users to "make sure the use of generative AI tools supports user and organizational needs."

2. Approved tools. List the tools staff may use and the data classification each one is cleared for. The federal guide's Secure principle is the test: "ensure that the infrastructure and tools are appropriate for the security classification of the information." Before a tool goes on the list, the same guide advises users to "understand how a system uses input data (for example, whether it's used as training data and whether it's accessible to suppliers)", and it asks institutions to have legal services, the chief security officer and the privacy team review a supplier's terms of use. A private business can assign those reviews to its own counsel, security lead and privacy officer.

3. Data classification: what may never be entered. This is the clause staff will read most. The federal guide is blunt: "Don't enter sensitive or personal information into any tools not managed by the GC." Read "not managed by the GC" as "not approved by us" when you adapt it. The privacy regulators' principles point the same way for public-facing tools: organizations should make sure the people using such a tool are informed of privacy risks and the mitigations available to them, "such as not entering personal information into a prompt, unless necessary." Write a short prohibited list (personal information, client confidential material, trade secrets, credentials) and tie each item to the tiers in your approved-tools list.

4. Human review and verification. Every output is a draft until a person checks it. The federal guide tells users, "Don't consider generated content as authoritative," and to "review it for factual and contextual accuracy." It adds a line worth copying verbatim: "Always review content generated by AI, even if the system seems to be reliable in providing accurate responses." NIST AI 600-1 names the underlying risk "confabulation", which it defines as "the production of confidently stated but erroneous or false content (known colloquially as 'hallucinations' or 'fabrications')". If your teams use AI on records about people, add the federal instruction to "verify personal information created using generative AI to make sure it is accurate, up to date and complete."

5. Disclosure and transparency. Decide when staff must say that AI was used. The federal guide's Transparent principle asks users to "identify content that has been produced using generative AI" and to "notify users that they are interacting with an AI tool." The privacy regulators go further for decisions: organizations should "clearly communicate to any affected party whether a generative AI tool will be used as part of a decision-making process." A practical clause sets three triggers: customer-facing chat, published content, and any decision about an individual.

6. Intellectual property and confidentiality. NIST AI 600-1 lists intellectual property as a generative AI risk, including the "eased exposure of trade secrets" and "plagiarism or illegal replication." The federal guide gives a checkable step: "Check whether system outputs are identical or substantially similar to copyright-protected material," and either give attribution or remove the material. Pair that with your existing confidentiality agreements, and state that prompts containing client material are disclosures under those agreements.

This is the section where counsel matters most, so keep it anchored to the text of the law and regulator guidance.

Accountability under [[PIPEDA|/terminology]]. The OPC's third-party guidance, dated September 10, 2026, states the accountability principle in its key points: "Organizations are responsible for personal information under their control," and this includes information "transferred to a third party for processing." A generative AI vendor that receives prompts containing personal information is a third party in that sense. The guidance asks organizations to identify whether any of the information is sensitive, to map how personal information flows between the organization, its clients and the provider, to account for where the information is stored, and to consider whether the provider intends to use the information for its own purposes. It also suggests writing those practices into the contract. Your policy can require that every approved tool has passed this assessment. As of November 2026, the OPC says it is accepting comments on this guidance until December 4, 2026, and will then evaluate whether any amendments are required.

The regulators' principles. The December 7, 2023 joint principles state that "accountability for decisions rests with the organization, and not with any kind of automated system." They also recommend that organizations "undertake assessments, such as Privacy Impact Assessments (PIAs) and/or Algorithmic Impact Assessments (AIAs)" and have "a clearly defined internal governance structure for privacy compliance."

Quebec: decisions based exclusively on automated processing. If you operate in Quebec and a decision about a person is made only by an automated system, section 12.1 of the Act respecting the protection of personal information in the private sector applies. It requires an enterprise to "inform the person concerned accordingly not later than at the time it informs the person of the decision," and, on request, to tell the person what personal information was used, "the reasons and the principal factors and parameters that led to the decision", and of their right to have the information corrected. The person must also be given the chance "to submit observations to a member of the personnel of the enterprise who is in a position to review the decision." The CAI's compliance aide-mémoire, dated February 8, 2023, lists respecting new information and transparency obligations among the obligations taking effect on September 22, 2023, and, among the action items to complete by that date, lists a mechanism to inform a person who is the subject of a decision based exclusively on automated processing. Section 90.1 of the Act lists failing to inform the person, or failing to give them the opportunity to submit observations, in contravention of section 12.1 among the failures for which a monetary administrative penalty may be imposed. According to section 90.12 of the Act, the maximum penalty is $50,000 for a natural person and, in all other cases, $10,000,000 or, if greater, 2% of worldwide turnover for the preceding fiscal year.

The policy clause itself can be short: no decision about an individual is made solely by an AI tool without the privacy officer's approval, and where one is approved, the notice and review steps above are built in before launch.

8. Security. The regulators' principles name the threats directly, asking all parties for "ongoing awareness of, and mitigations against, threats" such as prompt injection, model inversion and jailbreaking. NIST AI 600-1 lists information security among its generative AI risks. This post stays at the policy level; for the threat detail itself, see AI Security Threats Every Enterprise Should Prepare For.

9. Training. The federal guide's Educated principle asks users to "learn about the strengths, limitations and responsible use of the tools," and its best practices include learning about bias, diversity, inclusion and values and ethics. Make completion of training a condition of tool access, and record it. Practical prompt-writing skills also belong here; HCC's Prompt Engineering for Enterprise post covers that side.

10. Incident reporting. Give staff one channel for reporting a prompt that exposed data, an output that caused harm, or a tool behaving unexpectedly. Under PIPEDA, the OPC states that businesses must "report to the OPC any privacy breaches that pose a real risk of significant harm to an individual," notify affected individuals, and "keep breach records of all breaches of security safeguards for two years." Your AI incident process should feed that existing breach process rather than run beside it. The federal guide adds an operational step: "Document problems, pause deployment, and update the tool if performance levels are not being achieved."

11. Review cadence. Set a fixed review date and an owner. GOVERN 1.5, as set out in NIST AI 600-1, calls for ongoing monitoring and periodic review of the risk management process to be planned, with roles and responsibilities clearly defined, "including determining the frequency of periodic review." The sources themselves change: the federal guide says "it will be updated regularly to keep pace with regulatory and technological change," and its page was last modified on September 24, 2026. Re-check each source at every review.

For organizations that later want a formal management system around this policy, ISO/IEC 42001 is one reference point; ISO describes it as a standard that "specifies requirements for establishing, implementing, maintaining, and continually improving an Artificial Intelligence Management System (AIMS) within organizations."

Treat the eleven sections as a checklist. For each one, write the clause in your own words, keep the citation beside it in your working draft, name an owner, and send the draft to counsel. The citations make the review faster, because the reviewer can see where each rule came from and whether it applies to your business.

HCC's AI Governance & Compliance service can help you map these sources to your tools and data, draft the policy sections, and prepare the material for your legal review. HCC's Corporate AI Training service can help you build the staff training that sections 3, 4 and 9 depend on. If you are not sure where to start, the AI readiness assessment is a free first step.

Government of Canada — Guide on the use of generative artificial intelligence Office of the Privacy Commissioner of Canada — Principles for responsible, trustworthy and privacy-protective generative AI technologies Office of the Privacy Commissioner of Canada — Guidance on assessing third-party service providers Office of the Privacy Commissioner of Canada — Privacy Commissioner of Canada releases guidance for businesses working with third-party service providers Office of the Privacy Commissioner of Canada — What you need to know about mandatory reporting of breaches of security safeguards LégisQuébec — Act respecting the protection of personal information in the private sector (P-39.1) Commission d'accès à l'information du Québec — Utilisation et communication de renseignements personnels Commission d'accès à l'information du Québec — Principaux changements de la Loi 25 Commission d'accès à l'information du Québec — Vers la conformité à la Loi sur le privé (aide-mémoire) NIST — Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1 ISO — ISO/IEC 42001:2023 AI management systems