AI is already in your workplace. Do you have a policy?

It can make work faster and better, but can also expose confidential information and take unintended actions

AI is already in your workplace. Do you have a policy?
Geoffrey Lowe

Exclusive to Canadian HR Reporter from Rudner Law.

An employee asks AI to polish up an email. Another uploads a spreadsheet to save time. A manager turns on an automated meeting assistant without considering what it will record or where that recording will go.

None of them thinks they are creating a problem. They are trying to get their work done. They may be exposing their employer to liability.

That is precisely why employers need an AI policy.

Why have a policy?

AI can make work faster and better, but it can also produce convincing nonsense, expose confidential information, and take unintended actions. Banning its use is impractical and often pushes employees to use AI without oversight. The solution is policy: clear rules establish expectations before enthusiasm outstrips judgment.

Without guidance, employees independently decide which tools are acceptable, what data to share, and whether AI outputs require verification. Some may assume any accessible tool is company-approved.

A policy replaces assumptions with clear expectations, helping employees use AI safely. However, policy alone is no substitute for oversight. Stating that employees are "responsible" will not prevent unauthorized uploads or correct inaccurate answers. Employers need workable procedures, training, and designated personnel to keep rules current.

What should the policy contain?

A clear purpose and scope: The policy should define its core objective: encouraging responsible AI use rather than prohibiting technology or pursuing innovation for its own sake.

Clear definitions of terms like generative AI and large language models should be included. Specific tools should be listed in an appendix rather than the policy body, allowing the company to update tool lists without amending the policy itself.

Coverage must be comprehensive, addressing all AI applications, including:

  • AI built into email, browsers, office software, HR platforms and customer systems
  • Meeting assistants
  • Generated images and audio
  • AI agents that can access files, send messages or initiate transactions.

The policy must explicitly state that using personal accounts or devices for work does not exempt employees from these rules.

Approved tools and permitted uses: Instructions must be specific. Vague guidelines like "AI is allowed" or "use your judgment" are insufficient.

Define approved tools, authorized users, and permitted use cases, along with a process to request approval for new tools.

Approvals should specify exact tasks. For example, permission to use AI for marketing copy does not authorize uploading medical records or connecting tools to confidential repositories.

Rules about information: Employees need to know what they can enter into an AI system, what requires approval and what must stay out.

Specify restrictions regarding personal data, medical records, legal communications, financial data, passwords, trade secrets, and client information across typed prompts, uploaded files, recordings, and system integrations.

Rules must align with applicable privacy laws. Removing names from documents may not suffice if remaining context can still identify individuals.

Review of finished product: AI can sound confident while being completely wrong. A polished answer is not a verified answer.

Review standards should match risk levels. Routine internal work requires user self-review, while high-risk or external outputs require fact-checking, citation verification, and qualified approval.

For AI tools capable of taking actions or initiating transactions, authorization controls must be established prior to execution.

The policy should explicitly state that asking an AI tool to verify its own output is not independent verification. Reviews must evaluate completeness, bias, confidentiality, and accuracy, beyond basic grammar.

Training, reporting and consequences: Proper implementation requires practical training for employees and management guidance on oversight and approvals. Enforcing policy breaches is difficult without adequate training.

Training must detail reporting procedures for accidental data uploads or incorrect outputs and provide confidential channels to raise concerns.

While concerns will be handled discreetly, employers should avoid promising absolute confidentiality when investigation requires disclosure.

Consequences for policy breaches should distinguish between deliberate misuse and honest mistakes, avoiding automatic penalties that discourage prompt reporting.

Keep it practical and keep it current: Crucially, your AI policy must reflect how your organization actually works. A lengthy document that nobody understands, or a review process nobody can realistically follow, will not accomplish much.

A policy is also not meant to be static and can evolve over time through observation of the way that a Company’s workforce uses and works with developing technology. An employer should regularly review its existing policies in order to identify gaps and assign responsibility for updates as tools and workplace practices change.

Ultimately, the goal is not to prevent AI usage, but to guide employees on when to use it, when to avoid it, and when to seek approval.

Geoffrey Lowe is an associate lawyer at Rudner Law in Toronto. He can be reached at (416) 864-8500 or [email protected].

Latest stories