AI and Intellectual Property in Canada

Canada's intellectual property statutes — the Copyright Act, R.S.C. 1985, c. C-42, and the Patent Act, R.S.C. 1985, c. P-4 — were written with human creators and inventors in mind. Neither statute expressly contemplates AI as an author or inventor. As AI systems become capable of generating creative content and inventing novel technical solutions, this assumption creates significant gaps in IP protection.

The central challenge for Canadian businesses is this: if your AI system generates a valuable work of art, a piece of software code, or a novel product design, you may not automatically own — or be able to protect — that output under existing law. Understanding these limitations is essential for structuring AI businesses, negotiating contracts, and protecting competitive advantages.

This entry addresses three IP categories: copyright (for creative outputs), patents (for inventions), and trade secrets (for AI models themselves).

The human authorship requirement: Copyright under the Copyright Act subsists in original works of authorship. Canadian courts and the Canadian Intellectual Property Office (CIPO) have consistently interpreted 'author' to mean a human being. This interpretation flows from the Act's text (which uses terms like 'author,' 'maker,' and 'performer' in human contexts) and from the foundational principle that copyright rewards human creative effort.

CIPO's current position: CIPO has stated that it will not register copyright in works generated entirely by AI without human authorship. Works generated purely by AI — without meaningful creative selection, arrangement, or expression by a human — are not eligible for copyright protection in Canada.

The authorship spectrum: In practice, AI is rarely used entirely autonomously — most AI-generated content involves some degree of human direction. The question is how much human contribution is sufficient to attract copyright:

• Minimal human input (simple prompts): Entering a text prompt into an AI image generator (e.g., 'a cat sitting on a moon') and accepting the first output is unlikely to constitute sufficient authorship. The creative choices are made by the AI, not the human.
• Substantial human direction: Using AI as a tool where the human makes significant creative choices — selecting from multiple AI outputs, substantially editing or combining AI outputs, providing detailed creative direction that shapes the work's expression — is more likely to attract copyright in the human's contribution.
• Human-authored, AI-assisted works: Works where a human author does the primary creative work and uses AI only as an auxiliary tool (grammar checking, image enhancement, reference generation) clearly attract copyright in the human author.

Implications for AI businesses: If your AI product generates creative outputs (marketing copy, images, music, code), those outputs may not be protected by copyright. Competitors could reproduce them freely. This means: 1. Contracts should be explicit about whether the AI provider claims any IP in outputs, and whether the client can protect the outputs (through sufficient human editing and authorship) 2. Business models depending on exclusive ownership of AI-generated content are legally fragile in Canada 3. Businesses generating large volumes of AI content may benefit from ensuring human editors make meaningful creative contributions to each significant output

Copyright in AI training data: Using copyright-protected works to train an AI model without a licence may constitute infringement of the right of reproduction under the Copyright Act. The 'fair dealing' exception in the Act (s. 29) permits use of copyright-protected works for purposes such as research, education, and news reporting — but commercial AI training is unlikely to qualify. This issue is pending before courts in Canada and internationally.

The inventor requirement: The Patent Act, R.S.C. 1985, c. P-4, requires that a patent application name the 'inventor' — the person who devised the invention. CIPO's current position is that an AI system cannot be named as an inventor on a Canadian patent. An AI-generated invention without a human inventor would not be patentable.

The Thaler v. Canada decision: This position was confirmed when Stephen Thaler's patent applications naming the AI system DABUS as the sole inventor were refused by CIPO. CIPO's refusal was upheld in Federal Court proceedings. Canada joined the UK, the US, and the EU in concluding that inventorship under existing law requires a natural person.

Human inventor, AI-assisted invention: Many patent applications involve inventions made by humans using AI tools (AI-assisted drug discovery, AI-assisted circuit design). In these cases, the human inventors — not the AI — are named on the patent. The patent system can accommodate AI-assisted inventions as long as humans remain the inventive minds.

Practical implications: If your business uses AI to generate or identify innovations: - Ensure human researchers or engineers are meaningfully engaged in the inventive process, not merely observing AI output - Document human inventive contributions carefully (lab notebooks, design records) to support inventorship in any future patent application - Be aware that purely AI-generated inventions cannot currently be patented in Canada, creating a window for competitors to use the same AI to arrive at the same 'invention'

Legislative reform: CIPO has indicated it is monitoring international developments and may revisit the inventorship question as AI capabilities evolve. The World Intellectual Property Organization (WIPO) has convened discussions on AI and IP. Businesses operating in cutting-edge AI fields should monitor these developments.

Given the limitations of copyright and patent protection for AI, trade secrets are often the most practical IP protection mechanism for proprietary AI models, training data, and algorithms.

What is a trade secret? In Ontario, there is no standalone trade secrets statute. Trade secret protection arises from common law (breach of confidence), employment law (implied duties of confidentiality), and contractual confidentiality obligations. A trade secret is information that: 1. Has commercial value because it is not generally known 2. Is subject to reasonable steps by the holder to keep it secret 3. Has been (or is at risk of being) disclosed without authorization

What can be protected as a trade secret in AI: - The model architecture and weights of a trained AI system - Proprietary training datasets (particularly curated, labelled datasets) - Training methodologies, hyperparameter choices, and fine-tuning approaches - Inference pipelines and post-processing logic - Benchmark performance data and evaluation frameworks

How to maintain trade secret protection: - Require all employees and contractors to sign robust confidentiality and non-disclosure agreements (NDAs) - Limit access to AI systems and training data to those who need it - Implement technical security measures (access logs, encryption, data compartmentalization) - Include trade secret protection obligations in any API or licensing agreement with clients - Conduct exit interviews with departing employees to remind them of ongoing confidentiality obligations

Limitation of trade secret protection: Unlike patents, trade secrets provide no protection against independent development. If a competitor independently develops the same AI model without using your confidential information, they are free to use it. Additionally, trade secrets can be lost permanently if confidentiality is breached — there is no 'cure' equivalent to patent re-issuance.

Regardless of the type of AI-related IP being created, ensuring that the IP is properly owned by the business entity — not by individual employees, contractors, or founders — is critical.

Employee-created IP: Under common law and most employment agreements, IP created by an employee in the course of their employment belongs to the employer. However, this implied assignment may not extend to IP created outside normal working hours or in areas outside the employee's job description. Ontario employers should use explicit IP assignment clauses in employment agreements to eliminate ambiguity.

Contractor-created IP: Unlike employees, independent contractors retain ownership of IP they create unless they explicitly assign it in writing. This is a critical trap for AI businesses that use contractors to develop models, write training code, or create datasets. Every contractor should sign an agreement containing an express IP assignment before commencing work.

Founder IP: Where a company is founded by developers who created early AI models or codebase before incorporation, the corporation must obtain a formal IP assignment from each founder. This assignment should be documented in writing and reflected in the minute book. Investors will conduct due diligence on this point.

Using third-party AI tools: If employees use third-party AI services (OpenAI, Anthropic, Google, etc.) to generate content or code in the course of work, review the AI provider's terms of service. Many providers assign outputs to the user, but with conditions. Some prohibit certain uses or include licence restrictions that may affect the business's IP chain of title.

Ontario AI businesses often monetize their technology through licensing rather than outright sale. Key licensing considerations for AI IP:

Scope of rights: AI licences should clearly define: the permitted uses of the AI system or outputs, any restrictions on field of use, territory, or sublicensing, and whether the client may use the AI's outputs commercially.

Model access vs. model ownership: Most AI companies license access to their model via API rather than distributing the model weights. This protects the trade secret in the model while providing commercial access. Ensure the API licence clearly prohibits reverse engineering, model extraction attacks, and unauthorized reproduction of the model.

Output licensing: If the licensor claims IP in AI outputs, the licence should address what rights the client has in those outputs and whether the client can use them to create derivative works or train competing models.

Open source AI models: Some AI businesses build on open source foundation models (e.g., Meta's Llama family, various open source models on Hugging Face). The licence terms of the underlying model apply to all downstream uses. Some 'open' AI licences prohibit commercial use or require attribution — always review the actual licence terms before building a commercial product on an open source AI model.

Canadian IP law presents significant challenges for AI businesses. Copyright may not protect purely AI-generated content, patents cannot be granted for AI-invented inventions without a human inventor, and trade secret protection — while practical — offers no protection against independent development. These gaps require deliberate business and legal planning.

Ontario AI businesses should: ensure human creative involvement in any content they want to own via copyright, document human inventive contributions to AI-assisted innovations, implement robust trade secret protections for proprietary models and data, use explicit IP assignment agreements with all employees and contractors, and monitor Canadian legislative reform in this rapidly evolving area.