With AI filtering out jobseekers, expert calls for more greater transparency
Ontario's job posting transparency law requires that companies inform candidates when artificial intelligence (AI) is used in hiring — but it doesn't address what happens when the system gets it wrong, says one expert.
HR professionals in Ontario and across Canada should use this moment to audit not just whether their job postings disclose AI use, but whether their hiring workflows give wrongly screened-out candidates a way to be reconsidered, says Gleb Tsipursky, a behavioural scientist and CEO of consultancy Disaster Avoidance Experts.
Since Jan. 1, 2026, Ontario employers with 25 or more employees have been required to disclose in publicly advertised job postings when AI is used to screen, assess or select applicants, under section 8.4 of the Employment Standards Act, 2000, introduced through the Working for Workers Four Act. The obligation applies regardless of whether AI plays a small or central role in a hiring decision.
The share of Ontario job postings referencing AI-related terms rose from 9% in October 2025 to 28% in May 2026, according to a recent report.
But disclosure only answers one question for candidates about AI being involved, says Gleb Tsipursky, author of The Psychology of AI Adoption at Work: From Resistance to Results,
"It does not answer the harder question: 'What happens when the system gets me wrong?'"
Bias risk when data or instructions go wrong
Tsipursky points to a well-known example of AI bias in hiring.
"AI has been used in employment evaluation for a while, and there have been a number of cases where AI has been shown to be biased because if you put garbage data in, you'll get garbage data out," he says. "Amazon had a well-known case study where it trained an AI algorithm on its past successful employees, and it had an inherent bias in hiring white male workers. So the AI tool that it created was rejecting people who weren't white and male, which obviously was not good."
He says the underlying issue isn't the technology itself. "It's about the data or the instructions," Tsipursky says, adding that when a tool isn't given the right data or the right instructions, "the tool will not be making the right determinations" and may flag candidates who are a good fit as a bad fit.
His recommendation: give rejected candidates a chance to respond before a decision becomes final.
“What I suggest is that the AI tool should explain why somebody seems like they don't fit a role, and give that person a chance to respond and address any problematic instructions or data that might have been fed into the tool."
Tsipursky adds that response could then form "part of the evidence submission that would improve their likelihood of fitting the role."
