For employers, executives and professionals, refusing to use AI is no longer prudence. It is commercial self-sabotage. AI can accelerate research, improve drafting, expose weaknesses in an argument and reduce the time required to make difficult decisions. But its ease of use has created a dangerous illusion: that typing into an AI platform is the digital equivalent of thinking out loud.
It is not.
The prompts an employer enters, the documents it uploads and the answers it receives may become evidence in litigation. Worse, inserting legal advice or confidential communications into an open AI platform may jeopardize solicitor-client privilege — the privilege courts afford to communications between lawyer and client to retain its confidentiality and avoid disclosure to the opposing side and to the court.
The seemingly private exchange taking place on a laptop today could become the most damaging exhibit in a courtroom several years from now.
Two risks deserve particular attention:
the inadvertent waiver of solicitor-client privilege; and
the production of AI prompts and outputs during litigation.
Solicitor-client privilege is among the most jealously protected principles in Canadian law. It permits clients to speak candidly with their lawyers, secure in the knowledge that their communications will remain confidential.
But privilege depends upon confidentiality. When protected information is voluntarily disclosed to an outside party, the privilege is waived.
The problems often arise innocently. A client receives advice from counsel and wants a second opinion. Or it wants an AI tool to simplify a legal memorandum, challenge the lawyer's analysis or suggest a better settlement strategy.
To obtain a useful response, the client supplies the facts. They may quote from their lawyer's email, upload correspondence, describe settlement discussions or identify perceived weaknesses in the case.
At that point, the user may have disclosed precisely the information the privilege was intended to protect.
Canadian courts have not yet provided guidance on whether using a AI in this fashion amounts to waiver. But a recent U.S. decision offers a warning.
In that case, a litigant entered information received from his lawyers into an AI tool, used it to generate documents and then provided those documents to counsel for the litigation. The court concluded that disclosure to an open AI platform amounted to disclosure to a third party and that privilege had therefore been lost.
American privilege law is not identical to Canadian law, and a U.S. decision does not bind a Canadian court. But that is scant solace. No employer or executive should aspire to become the Canadian test case.
The consequences can be severe. Advice about legal exposure, litigation strategy, settlement ranges and the strengths or weaknesses of a case may become disclosable to an opposing party.
A few careless minutes with an AI platform might compromise months or years of protected communications.
For employers, the danger extends well beyond privilege. Even where no privileged communication is involved, AI prompts and outputs may become relevant and producible in litigation.
Employers are increasingly turning to AI for assistance with terminations, performance management, harassment complaints, disability accommodation, restructuring and employee discipline.
The appeal is obvious. Why pay a lawyer when a platform can produce an immediate answer at little or no cost?
Why? Because the "free" answer may later become an extremely expensive exhibit.
Employment cases frequently turn on intent. Courts and tribunals examine what the employer knew, when it knew it and why it acted as it did. They scrutinize internal emails, draft memoranda, meeting notes and communications among decision-makers.
An AI prompt may offer a very candid record.
Consider a company contemplating the dismissal of a long-service employee. A manager might ask:
"How do we terminate this employee while paying the least possible severance?"
Or:
"How can we avoid a human rights complaint?"
Or:
"Can we eliminate an employee's position while she is on disability leave?"
The manager may regard those questions as preliminary research. Opposing counsel may characterize them very differently. And a court may find the question damning.
A prompt framed carelessly could be used as evidence of bad faith, discriminatory intent, retaliation or a deliberate attempt to evade statutory obligations.
Even if that characterization is unfair, the employer will have surrendered enormous negotiating leverage. Instead of litigating the merits of the dismissal, it may be tasked with explaining an ugly-looking prompt casually drafted months earlier by someone who never expected anyone else to read it.
The AI output can create further problems. An employer may rely on an answer that is legally wrong, based on advice from another jurisdiction or oblivious to facts that would have changed the analysis.
AI writes with confidence even when it is mistaken. That combination is dangerous in employment law, where the result frequently depends on the wording of a contract, the employee's jurisdiction, the employer's prior conduct, the most recent caselaw and the precise chronology of events.
None of this means that businesses should stop using AI. That would be both unrealistic and commercially foolish.
It means they must govern its use with the same seriousness they apply to email, document retention, cybersecurity and legal communications.
Employers should adopt clear rules prohibiting employees from entering privileged advice, confidential legal correspondence, personal employee information or sensitive business records into unapproved AI platforms.
They should also require employees to assume that every prompt may one day be reviewed by a judge, regulator, investigator or opposing lawyer.
Would I be comfortable having this prompt read aloud in court?
Would I be prepared to explain why I asked the question in this way?
Does this prompt disclose confidential, privileged or personal information?
Am I using AI to assist judgment, or to replace the professional advice the situation actually requires?
AI can be an extraordinarily useful tool. But it is not a confessional, and it is not a privileged conversation simply because no one else is in the room.
The most damaging evidence in the next generation of employment cases may not be found in an email, text message or board memorandum.
It may be found in a prompt someone wrongly assumed no one would ever see.
Howard Levitt is senior partner of Levitt LLP, employment and labour lawyers with offices in Ontario, Alberta and British Columbia. He practises employment law in all provinces and is the author of six books, including the Law of Dismissal in Canada. Jeffrey Buchan is an associate at Levitt LLP.