Back to Blog
AI Adoption8 min read

Who Owns What Your AI Creates? Canada's IP Gap and What SMEs Need in Their Contracts Now

By Anton Kuznetsov

Most Canadian SME owners have not thought carefully about who owns the AI-generated content, code, and analysis their teams produce every day. That oversight is becoming a liability.

Canadian IP law has not caught up to generative AI. The Copyright Act has no provision addressing AI-generated works. The Patent Act has no AI inventor clause. The Office of the Privacy Commissioner's guidance addresses data, not outputs. And CIPO rewrote its entire approach to software and AI patents in March 2026 — without anyone in most SMEs' legal departments noticing, because most SMEs do not have legal departments.

This post covers the three IP questions every Canadian business using AI needs to answer, and the contractual steps that answer them.

The Copyright Gap: Humans Still Required

Canada's Copyright Act protects original works created by human authors. The Act has not been amended to address AI-generated works, and no legislative proposal to do so has been tabled as of this writing.

Lexpert's analysis of the question summarizes the current state plainly: Canadian law offers no clear answer as to who owns the inputs or outputs resulting from the use of AI. Courts have not settled the question. CIPO has not issued copyright guidance on AI-generated works. The government's June 2026 AI for All strategy contains no legislative proposal to fill the gap.

What does this mean practically? If your marketing team asks Claude to write a campaign and the team's input is a minimal prompt — "write a product description for X" — the resulting content may not attract copyright protection at all. You cannot own what has no owner. A competitor could reproduce it, build on it, or use it without consequence.

The minimum threshold for human authorship — enough human creative expression to qualify — is contested in both Canadian and US courts. One practitioner summarized the problem directly: "If you prompt it with a very minimal prompt and it outputs something that you end up using, the answer is not you."

The practical response is not to avoid AI-generated content. It is to structure AI use so that human authorship is embedded in the output: editing, reorganizing, selecting, combining, and annotating AI-generated drafts rather than accepting first outputs unchanged. Document that human contribution. The more a person shapes the final product, the stronger the claim to copyright.

The Patent Question: CIPO's March 2026 Rewrite

If your company is building AI-assisted products or services that could be patented, the Canadian Intellectual Property Office changed the rules in March 2026.

CIPO's new Practice Notice, issued March 24, 2026, replaces the November 2020 notice that most patent agents had been working under. The new notice drops the "actual invention" step that had been part of the 2020 framework, folds that analysis into purposive construction, and places greater weight on physicality as the key criterion for patentability.

The core test now is whether a claimed AI or software invention has real-world physical effects — meaning it improves a computer's own functioning, operates on physical processes, or produces tangible physical results. The notice includes Canada's first worked example addressing machine learning: a neural network trained on historical weather, irrigation, and crop yield data to generate irrigation recommendations. That example helps define where the line sits.

What has not changed: software per se remains non-patentable in Canada. Claims must refer to tangible media or a physical system. The notice does not address quantum computing specifically. And the new framework applies retroactively: applications under examination are assessed under the 2026 notice, not the 2020 one.

DLA Piper and Smart Biggar both recommend that any organization with pending Canadian patent applications covering AI or machine learning review those applications against the new guidance, and that companies planning to file AI-related patents in Canada understand the physicality requirement before drafting claims.

For most Canadian SMEs, this is relevant in two scenarios: if you have built or are building an AI-assisted product that might be patentable, and if you are licensing technology that another party claims is patented. Either way, the March 2026 notice changed the landscape.

The Contractor Problem: Canada Has No Work-for-Hire Doctrine

When a Canadian business hires an independent contractor to build software, write content, or develop AI tools, a critical assumption breaks down. Under US law, a "work for hire" agreement can automatically vest copyright in the client. Canada does not recognize this doctrine for independent contractors.

Under the Canadian Copyright Act, an independent contractor is both the author and first owner of the work they create. Ownership passes to the client only through a valid written assignment. If that assignment is missing or unclear, the contractor owns the IP — and can license it to others, build competing products from it, or refuse to transfer it.

A 2026 Mondaq analysis of AI technology agreements notes that many Canadian organizations are relying on US-style contractor agreements that do not include Canadian-specific IP assignment language. The article explicitly warns that "a contract stating that you own IP rights does not mean that you actually do, because IP laws govern this rather than contracts." The assignment must be valid under Canadian law, not just present in a document.

This applies with particular force to AI-related work:

  • If you hire a contractor to build a Claude-powered agent workflow for your business, who owns the prompts, the retrieval logic, the fine-tuned configurations?
  • If your developer builds an internal AI tool on company time but under an independent contractor arrangement, whose IP is it?
  • If an AI vendor's model outputs are incorporated into something your contractor builds, does the ownership chain still hold?

The answer to each depends on the specific assignment language in the contract. Where that language is absent or vague, ownership defaults to the contractor or the vendor — not to you.

AI Vendor Contracts: The IP Risks Most Businesses Miss

Your AI vendor's terms of service determine whether the AI outputs your team generates could be claimed by a third party — either the vendor or the data sources the model was trained on.

Bennett Jones' April 2026 analysis of AI use, confidentiality, and IP exposure identifies several contractual gaps that Canadian businesses frequently miss:

Training data risk. Many AI models were trained on data that may include copyrighted material. If a vendor cannot warrant that its training data was appropriately licensed, outputs derived from that training may carry infringement exposure. This exposure passes through to you as the user.

Output licensing. Some vendor agreements assert rights over outputs generated through their platform. Others disclaim all warranty over whether outputs infringe third-party IP. Neither position necessarily protects you.

Secondary use of your data. Separate from the IP question, vendor agreements that permit training on user prompts and inputs raise both PIPEDA accountability issues and IP concerns: your proprietary workflows, formulations, or analysis techniques could feed into the vendor's model and benefit competitors.

The Mondaq analysis on negotiating technology agreements recommends that Canadian organizations require AI vendors to represent that their models do not infringe third-party IP rights and that their outputs will not expose users to infringement liability. These representations should be explicit, not implied.

For enterprise AI platforms — Claude Enterprise, Microsoft 365 Copilot, AWS Bedrock — vendor agreements are structured and obtainable. The IP warranties and data processing terms are documented. For smaller or newer AI tools, the absence of these warranties is itself a risk signal. If a vendor cannot answer the question directly, that answers the question.

What Belongs in Your AI Governance Policy

The Canadian Lawyer Magazine's 2026 analysis of AI governance gaps quotes BCF's technology law practice as identifying IP policy as one of the three most common missing elements in Canadian organizations' AI governance — alongside data privacy controls and acceptable use guidelines.

The minimum additions an AI policy needs to address IP:

1. Authorship documentation. Require employees to document the human contribution to AI-assisted outputs that the organization intends to claim as its own: editing logs, iterative revision records, creative decision notes. This is not bureaucracy — it is the evidence that establishes copyright.

2. Contractor IP assignment checklist. Every new contractor engagement involving AI development should use a Canadian-specific assignment clause reviewed by a lawyer familiar with copyright assignment requirements. US-style boilerplate is not sufficient.

3. Vendor IP warranty requirement. Add a checkpoint to tool onboarding: before any AI tool gets access to proprietary data or workflows, require a written representation that outputs do not infringe third-party IP, and check whether the Data Processing Agreement addresses secondary use of your inputs.

4. AI-generated prior art awareness. If your company has pending or planned patent applications, brief your patent agent on what AI tools your R&D team is using. AI-generated descriptions of solutions — in emails, shared documents, or vendor pitches — can constitute prior art that is searchable and datestamped, potentially undermining your patent claims.


None of this requires a legal department. It requires knowing the questions and asking them before a dispute surfaces them for you.


Sources


Cloud Forces' AI Advisory service includes a structured review of your AI governance program — covering acceptable use, vendor due diligence, and the IP and contractual gaps that most organizations discover only after a dispute. As an AWS Consulting Partner since 2019 and a Microsoft Solutions Partner since 2022, we bring the infrastructure and vendor context to make your governance program practical rather than theoretical. If your team is deploying AI tools and you have not reviewed your contractor agreements or vendor terms through an IP lens, book a consultation before the gap becomes a problem.

Anton Kuznetsov
Founder & Principal Engineer

Anton Kuznetsov is the founder and principal engineer of Cloud Forces, the Toronto firm he started in 2018 to make AI and cloud practical for Canadian SMEs. He leads Cloud Forces’ AI advisory and Claude deployment work and oversees the secure cloud platforms the firm runs for its clients.

Ready to bring AI to your business?

Book a free AI Readiness Consultation — no commitment required.

Book Free Consultation