‘Long story short, HR needs to be at the centre of AI governance’: researcher warns Canadian employers to get governance right, right now
The continuing saga around Sam Altman’s contested control of OpenAI puts a spotlight on the huge issue of ethics when it comes to artificial intelligence.
A recent The New Yorker investigation by Ronan Farrow describes secret board memos alleging Altman “misrepresented facts” to his own team and directors who worry he should not “have his finger on the button” of powerful AI systems.
OpenAI’s former chief scientist Ilya Sutskever compiled the memos, according to the report, alleging that Altman lied to directors about internal safety protocols and showed “a consistent pattern” of deception.
For Canadian employers bringing AI into hiring, performance management and everyday work, safety and ethical use is less about the tools themselves and more about employees trusting leaders to use it responsibly, according to one expert.
Breaking down the tech-guru narrative
Shion Guha, assistant professor of information and computer science at the University of Toronto and director of the Human Centred Data Science Lab, is quick to address the outsized role that tech CEOs have in shaping public narratives.
“Unfortunately, the media has an issue where any random comment that these folks make is kind of taken as the as the gospel truth,” he says, pointing out that the tech mega-companies such as OpenAI, Anthropic and Microsoft aren’t posting big profits from AI.
“None of them are actually profitable. So, Sam Altman needs to make money. That’s why he makes comments. The more ludicrous, the better, because that’s attention- and eye-catching.”
This pattern is repeated in other workplaces as leaders push the use of AI and other tech advancements as the answer to budgetary prayers but lack the governance and training to back it up – bringing up big safety and ethical risks.
“Leadership integrity matters a lot more in these contexts,” Guha says, “because AI governance fails long before any particular AI model will fail.”
Why Canadian employers can’t ignore AI governance
Guha’s research focuses on how data, measurement and organizational practice interact when institutions adopt algorithmic systems.
In that work, he sees a common Canadian problem that’s spread across all sectors.
“When big tech companies make AI and then they sell these products to other companies in our economy, one of the most important things we have to remember is that individual companies are responsible for their internal AI governance regimes,” Guha says.
“At least in the Canadian context, very, very few companies actually have comprehensive AI governance mechanisms or regimes… It’s kind of being done on a very ad hoc perspective.”
What good AI governance looks like
Guha offers a straightforward test for employers trying to move beyond ad hoc pilots into complete governance regimes for AI: documented responsibility, escalation paths or contestability, and authority and discretion.
For HR leaders, this three-part test can be translated directly into policies and processes, he says.
Documented responsibility means naming which role – not just which department – is accountable for each AI system that touches employees, from applicant tracking to performance analytics. Escalation paths mean baking in ways for employees to question AI-influenced decisions, and making sure managers know those channels exist.
Authority and discretion mean identifying the senior leaders who can override or pause a system when something looks wrong, rather than leaving line managers to “make do” with outputs they distrust or systems that aren’t working.
In practice, that means someone with real clout can intervene when AI systems go off track.
“Who has the authority to pause the system or shut the system down when it starts producing unsafe or unreliable outputs?” Guha asks.
“A rank-and-file employee is not going to have the authority, they’ll be like, ‘Hey, my boss said I need to use AI, so I’m gonna throw everything into ChatGPT’ … right now in Canada, there’s very, very few places where this exists in conjunction with the deployment of AI.”
HR at the centre of AI governance talks
Too often, Guha says, AI governance is treated as a technical problem for IT teams alone, leaving a gap where employment law and workplace culture should be. When HR is only invited in at the end, he explains, policies don’t reflect real world concerns such as morale, trust and accommodation duties.
“Long story short, HR needs to be at the center of AI governance. As AI governance is being designed in any organization, HR needs to be on that ground floor,” he says.
“Usually what happens … there’s some AI steering committee, and they come up with some type of a plan, and then they call HR and they say ‘Okay, HR, now we have a policy, and HR, you need to enforce the policy.’ And that usually doesn’t work.”
To see long-term ethical and successful AI adoption that doesn’t upend employee relations, organizations need to include HR throughout the process.
“The whole thing needs to be co-designed with HR, because there’s things around employee engagement, and anxieties and fairness concerns and accommodations, as well as things like reputational risk,” Guha explains.
“These have legal liabilities, and HR is very aware of that, this is what they deal with at the ground floor.”
From pet projects to real accountability
Even when organizations do set up “ethical AI” teams, Guha warns they are often underpowered and symbolic. He points to former students who have helped start such groups at major Canadian firms, only to find that decision-makingg remains concentrated elsewhere.
For HR, that means good intentions can sit in a corner of the chart as the “pet project” of one leader or other, while, in actuality, day-to-day AI decisions continue unchecked.
“Maybe it’s the CFO’s pet project, but the CEO doesn’t buy into it,” or an EVP’s project that is largely ignored, Guha says.
“That’s the issue. Governance has to be company-wide. It can’t be a ‘special thing’.”
Proper AI governance starts with a more democratic design of org governance itself. Rather than a small group imposing rules from the top, Guha says involving those who will live with AI systems day-to-day is crucial, including HR, frontline staff and union representatives where applicable.
“A lot of the time, people will come up with stuff, and then it becomes a report, and nothing ever gets implemented,” Guha says.
“You need one strong group that has a representation of all of the major stakeholders across the company, to co-design it from the bottom up. It can’t be top-down. It can’t be something that a leader says: ‘This is how we should do governance.’ That’s very silly. Governance comes from the bottom up.”
