'So far, the courts have not been very sympathetic towards companies utilizing AI in place of people to make decisions'
Canadian HR professionals are increasingly using generative artificial intelligence (AI) for a variety of tasks, such as onboarding and performance management.
But what about legal documents such as employment contracts, employee handbooks and workplace policies? While these tools offer efficiency, they also introduce new risks and challenges, according to two employment lawyers providing insight into the benefits and risks of relying on AI for legal HR tasks.
“I would be very, very careful using an AI tool to draft a termination letter for you, because it's not accurate most of the time — I wouldn't even say 50% of the time,” says Sunira Chaudhri, litigator and founder of Workly Law.
“Organizations as a whole need to be really careful about using AI to… place way more responsibility on HR professionals thinking, ‘Well, we need a smaller HR team because we now have all these cool AI tools to do a lot of our drafting.’ That could be short-sighted, because… there could be a big mistake.”
Balancing efficiency and risk with AI
AI’s appeal in HR is clear: it can help manage large volumes of information and speed up drafting.
“From the HR lens, we are certainly seeing more and more employer clients looking for more efficient ways, particularly when it comes to hiring,” says Chaudhri.
However, she cautions that legal risks begin even with seemingly simple tasks such as job ads, as the platform may not be aware of the changing legislation.
“We know laws have changed now with respect to what must be included in a job posting. For example, there are laws with respect to salary ranges. So, there is a bit of additional scrutiny being placed on job postings and legal considerations that AI tools may not be able to grasp,” she says.
Plus, there’s been plenty of coverage around the risks of bias in screening tools, says Chaudhri.
“There’s been reports that AI could be skewed to certain genders, or those that are younger versus older,” she says. “You certainly have to be really careful about using AI tools for screening, to the extent that that tool might, on some discriminatory ground, screen out one class of applicants over another.”
Legal accuracy and risk of ‘hallucinations’
AI’s tendency to make errors is a big concern, such as not staying on top of case law or legislative changes, according to Jeffrey Adams, labour, employment and human rights lawyer at Spring Law in Toronto.
“The big errors I see are often when the law changes and AI doesn't react as fast as a lawyer could,” he says, such as AI citing a decision, not realizing it was overturned on appeal.
Adams gives another recent example, when the Canada Labour Code changed in 2024: “They dramatically increased the separation payments on termination — it went from two weeks to graduated, much more similar to the ESA — and AI didn't update that [for a client]... They got their answer from ChatGPT and a year ago, it would have been right, but the law changed.”
Of course, there’s also the huge risk of ‘hallucinations,’ where AI just makes up facts.
Recently, a lawyer in an Ontario court was sanctioned for using the tech to create a legal brief, which cited non-existent legal cases, warns Chaudhri.
“AI as it stands, has not accurately, in many recent instances, interpreted the law well so I don't think you can leave an AI tool to, for example, ensure that your termination letter is on side of employment legislation. It's just not there yet.”
Who looks at the final draft?
And if HR is using AI to ask questions or prepare a termination letter, that could up in a litigation file some day, she says.
“That's not going to remove the liability from the organization or even from the HR person, but it's going to place into question: ‘Well, was AI the last set of eyes that looked at the document? Or was it you?’
“And where we don't have a good answer to that question, that's where I think we're going to see cases turn because, so far, the courts have not been very sympathetic towards companies that have been utilizing AI in place of people to make decisions, to draft documents.”
While the use of AI could make things more efficient and faster, and it could get things right 90% of the time, termination letters are always scrutinized in wrongful dismissal cases, says Chaudhri.
“They are placed through the ringer — judges take a magnifying glass to each word, and being 90% right in a termination letter is not enough… and you definitely don't want to be walking into a court saying, ‘I didn't draft this, my AI tool did.’”
Another big challenge for HR? AI-generated documents often lack the nuance needed for compliance with local laws.
An employer using AI to create one-size-fits-all employee handbooks, for example, may not appreciate the employment laws specific to Ontario compared to the rest of Canada and “will often run you afoul of local legislation,” she says.
Privacy and confidentiality concerns
Both lawyers also highlight the privacy risks of using AI when it comes to inputting confidential information from the company, employees or customers.
“Most employers know that employees are using AI to some degree, but most have not placed parameters around it. And I think it's largely because there remains a large misconception around what AI does, what the risks of it are,” says Chaudhri.
And while there might be closed AI systems, where the information isn’t necessarily fed back into the model at large, many employees are still using the public systems, she says.
Adams says he doesn’t trust any of the publicly available genAI models, especially the free ones: “There's no illusions about them learning from the data that is inputted,” he says, adding that reputable models can provide greater confidence.
“It still can't be bulletproof — I think you always have to be very cautious,” he says. “I scrub as much of a document as I can, I’m very careful.”
Working through strategies
Adams agrees that AI can be a useful starting point, especially for research and brainstorming.
“If you tell it to provide four different strategies of varying levels of aggression for this set of circumstances, so it's helping clients work through strategies to approach legal issues ... it's a really good starting point. And that's where I think it's most valuable.”
Adams admits he doesn’t use the technology much for drafting, but does use it for large data sets in culling information.
“You can upload a compendium and, say, ‘Find all of the instances where there's evidence of this issue,’ and it'll pull it all out for you and hyperlink to where in the compendium It's located,” he says.
“But you have to go to every single footnote and check because there's no shortcuts there. So, it helps with the drudge work but the real work is still the lawyer doing the proofing, making sure the citations are correct.”
Adams admits that he has clients who use AI, and then ask him for his opinion. He says he encourages them to use AI for research and to focus their questions.
“One of the prompts that I encourage clients to use all the time is just say, ‘Are you sure?’ to ChatGPT, and it'll proof itself sometimes. It's not foolproof, but [it’s about] asking that next level of questions, just to make sure.”
Need for human oversight with AI
Both lawyers stress that AI’s output must always be reviewed by a human expert.
“I don't send anything out just on blind faith, I look at AI as a good starting point,” says Adams.
“You can create policy, you can get it to draft policy, it'll be very good, but it's not covering all the bases. And that's where you make one little mistake in a policy, and you're in a bunch of human rights hot water. So, caution is the operative word.”
AI is great as long as it’s being used by someone “with the knowledge, the sophistication and the deep experience in the field to know when AI has got it a bit wrong,” says Chaudhri.
The human element is also important when it comes to the type of organization you are, and how that’s reflected in legal documents, she says.
“Are you an employer that wants to pay the Employment Standards Act minimums and that's it? Or are you looking for something a bit more generous and in what circumstances?
“Having that conversation about what the goals are of the organization, and making sure employment agreements, termination letters match that vision and that mandate, and then using the AI to get you there, I think is the most efficient way of doing it.”
Using AI as a tool, not a substitute
Offering reassurance, Chaudhri says that HR folks are never going to be replaced, and there’s no substitute for a great HR leader.
“But I think [it’s about] using AI as one of the tools in your toolkit, having good legal counsel, having a good AI or drafting tool, and creating updated policies and handbooks,” she says.
“When it comes to contracts and key legal documents, there's really no substitute for a great legal mind and a great HR leader mind coming together to make that work.”
