Plan sponsors face a growing automation bias risk, says ADP's Helena Almeida, as AI embeds itself in HR, payroll, and benefits workflows
Artificial intelligence is no longer a separate tool that sits on the sidelines of workplace operations. After all, it schedules interviews, drafts job descriptions, checks compliance on postings, and provides benefits information for employees navigating complicated eligibility rules with a single prompt.
Now, the question facing employers is no longer whether to use AI but who bears responsibility when it gets something wrong and whether anyone is still checking.
“One of the risks that I worry about most is not really just the AI getting something wrong, it's the human becoming too willing to assume that the AI is right,” said Helena Almeida, vice-president, managing counsel at ADP Canada. "Because AI can produce something instantaneously, and it's polished and confident, it makes the output feel really authoritative.”
Automation bias erodes human oversight
That authority is exactly the problem here as Almeida pointed to a pattern she calls the shift from reviewer to approver, a workplace dynamic where the speed and polish of AI output erodes the critical thinking it's supposed to support. The risk is also known as automation bias and it's compounding as AI moves deeper into decisions that affect people's livelihoods, from payroll calculations to performance reviews and compensation.
That spectrum - from information to action - is where governance needs to be calibrated, Almeida said, breaking down the work into four steps that she conceded sound basic but form the foundation of responsible AI use. First, organizations need to know where AI is being used, cataloging enterprise-approved tools as well as informal experimentation by managers and teams. Then, they need to classify the risk. Third, organizations need controls, like implementing policies on when AI can be used, human review requirements, vendor due diligence questions, and channels for employees to raise concerns. Finally, employers need to monitor AI use and its outputs.
"It's really important to make sure that we're designing the workflow, so people remain engaged and know when their judgment is expected. Otherwise, we fall into that risk of an automation bias that anything an AI tool says is accurate," she added.
Why trust is paramount for AI at work
Almeida believes that AI is only useful in a workplace if people trust the outcomes enough to act on them. That trust, she argued, depends on a handful of core questions: whether the output is accurate, whether employees know when AI is involved, whether there's room for human oversight, and whether steps have been taken to address bias.
Almeida said human oversight demands more scrutiny because AI is embedding itself deeper into everyday workplace functions from HR, recruiting, workforce management, benefits and those opportunities notably come with a governance challenge. The more seamless and persuasive the technology becomes, the easier it is for people to accept its outputs without scrutiny.
"This human in the loop isn't necessarily the same thing as meaningful human judgment. We want to make sure that the person reviewing the output really understands what the AI is doing, that they're able to question it, that they have the authority and ability to override it, and that they know that they ultimately own the decision," said Almeida. "These are important guardrails to make sure that the AI is used responsibly and that ultimately the employers and the employees can trust the outcome.”
Should HR leaders place complete trust in AI?
To that end, she encouraged organizations to reinforce with their teams that AI is a tool, not an authority, and that employees should feel empowered to question or override its outputs when something doesn't seem right.
Moreover, she suggests the goal isn't to resist the technology but to make sure its speed, scale, and analytical power don't outpace the judgment and accountability that workplace decisions require.
Most organizations and most individuals are still on the learning curve, figuring out how to use AI in ways that actually boost productivity rather than create new friction, Almeida said. The companies seeing real results, she noted, are the ones that have rethought their workflows from the ground up rather than layering AI onto legacy processes.
HR leaders lack AI ethics policy, still trusting of AI
According to ADP Canada's 2026 Workplace Trends Report, only 21 per cent of companies use AI for compliance tasks, and among those that do, 51 per cent express a strong trust in AI’s accuracy. Additionally, 70 per cent of Canadian businesses currently don’t have an AI ethics policy. While 46 per cent agree that ethical management of AI is a priority, only 22 per cent have taken the steps to do so.
Meanwhile, additional data found 88 per cent of organizations see value in pairing compliance expertise with AI, a figure Almeida believes reinforces the case for governance, not replacement. She suggests awareness and action are two different things, particularly as most organizations, she noted, haven't figured out how to operationalize responsible AI use, even as they recognize its importance.
"The fact that AI contributed to a decision doesn't make accountability disappear," she said. "An organization should be able to answer what role is the AI playing. Is it providing information, making a recommendation, or actually taking an action?"
