Does my startup need an AI use policy for employees?
Written by For Founders Law, a Toronto firm licensed by the Law Society of Ontario. About us
Short answer
No law requires one, but most startups should have one once anyone on the team uses AI tools with company or customer information. Pasting personal information or a customer's confidential data into an AI tool can breach PIPEDA's safeguarding principle, your NDAs and your customer contracts, depending on the tool's terms. A short policy that names approved tools, lists what can't be entered, and requires human review of output covers most of the risk.
Most teams start using AI tools long before anyone writes a rule about it. An AI use policy closes that gap with a page or two of practical instructions.
What are the legal risks without one?
| Risk | Where it comes from |
|---|---|
| Disclosing personal information | Under PIPEDA’s Schedule 1, you stay responsible for personal information you send to a service provider, and must protect it with safeguards suited to its sensitivity |
| Breaching customer contracts | Many SaaS agreements and DPAs limit which sub-processors can receive customer data. An unapproved AI tool may be one |
| Breaching NDAs | Confidential information shared with an AI vendor may be a disclosure the NDA doesn’t allow. See NDAs for startups |
| Losing trade secret protection | Secrecy has to be actively maintained. See how to protect trade secrets |
| Unclear IP in output | Copyright law hasn’t settled who, if anyone, owns purely AI-generated material. See who owns AI-generated content |
| Relying on wrong output | Output used in customer advice, code or contracts without review |
The answer often depends on the tool’s terms. Business and enterprise plans of many AI tools say inputs aren’t used for training and offer data processing terms; free consumer plans often say something different. Read the terms of each tool you approve.
What should the policy say?
Keep it short enough that people read it:
- Approved tools, with the account type (for example, the company’s business plan, not personal accounts).
- What never goes in: customer personal information, health information, passwords and keys, customer confidential data, and unreleased financials, unless the tool is approved for that data.
- Review: a person checks output before it goes to a customer, into production code, or into a contract.
- Code: check AI-suggested code for licence issues and security problems. See open-source licences in AI products.
- Disclosure: when to tell customers that AI was used.
- Who to ask, and how to request a new tool.
Does this affect employment agreements?
Your employment and contractor agreements should already say the company owns work product and that confidential information can’t be disclosed. Under the Copyright Act, employers generally own copyright in work employees create in the course of employment; contractors need a written assignment. See does my company own what employees create.
If you monitor how staff use AI tools and you have 25 or more employees in Ontario, your electronic monitoring policy should describe it. See what workplace policies an Ontario employer needs.
What to do next
We prepare employment agreements and contractor agreements with confidentiality and IP terms, and AI product terms for products you sell, at fixed fees on our fees page.
This guide is general information about Ontario and Canadian law, not legal advice for your situation. To talk through how your team uses AI, book a free 15-minute consult.
Sources
- PIPEDA, Schedule 1, Principles 4.1.3 and 4.7 (accountability for transferred information; safeguards)
- Office of the Privacy Commissioner of Canada: Principles for responsible, trustworthy and privacy-protective generative AI technologies
- Copyright Act, R.S.C. 1985, c. C-42, s. 13 (ownership of copyright)
- Ontario: Your guide to the Employment Standards Act, written policy on electronic monitoring of employees