The algorithm is not your lawyer: the limits (and dangers) of AI tools

In many HR departments across North America, artificial intelligence tools are now drafting termination letters, summarizing workplace investigations, estimating severance ranges. Photo by Getty Images

Article content

Artificial intelligence is rapidly embedding itself into corporate decision-making. In many HR departments across North America, AI tools are now drafting termination letters, summarizing workplace investigations, estimating severance ranges and even offering views on whether a dismissal complies with statutory requirements.

Article content

The efficiencies are obvious. The risks, less so.

Article content

Article content

Article content

AI can be a valuable administrative tool. But it is increasingly being used not merely to assist but to replace professional legal judgment. That shift should concern employers.

Article content

Article content

Modern AI systems can summarize legislation, reference case law and articulate legal tests with impressive apparent fluency. The output is polished, structured and often persuasive.

Article content

But that fluency creates a false sense of credibility — even certainty.

Article content

Employment law is among the most fact-specific areas of law. A termination clause that appears enforceable may be void because of subtle drafting defects. An investigation that follows procedural steps may nonetheless fail on fairness grounds. Constructive dismissal cases frequently turn on context, credibility, motive and evolving appellate interpretation.

Article content

AI tools cannot reliably:

Article content

Article content

The result is not necessarily an incorrect analysis, but it’s an incomplete one. And incomplete analyses in employment law can be very expensive.

Article content

Article content

The jurisdictional issue is particularly acute in Canada. Reasonable notice under Ontario common law often exceeds statutory minimums under the Employment Standards Act by significant margins. Human rights regimes vary by province and federally. Appellate courts regularly refine termination clauses enforceability.

Article content

Article content

AI systems trained on broad datasets may blend legal concepts across jurisdictions without clear differentiation. Analysis influenced by U.S. at-will principles is of little assistance to a Canadian employer, where dismissal without cause triggers substantial notice/severance obligations.

Article content

In addition, AI platforms typically do not have access to subscription-only legal databases. Recent appellate developments may not be reflected in their outputs. Employment law evolves quickly. Static training data does not.

Article content

There is also a governance issue.

Article content

When employers input detailed fact patterns into public AI platforms, they may be exposing confidential employee information. More significantly, they may compromise solicitor-client privilege.