Are we getting workplace AI regulation all wrong?

While existing policy might look good on paper, it is too often based on headlines and edge cases, and is therefore impractical and potentially dangerous for the average worker, new research has found.

Commissioned by the Shop, Distributive and Allied Employees Association (SDA), a new report from the John Curtin Research Centre urges workplaces to consult with their employees as soon as possible regarding AI implementation.

The report, titled For All of Us: Making Artificial Intelligence Work for Working People, highlighted “excessive surveillance, unrealistic productivity targets and work intensification” as potential consequences of unchecked AI adoption by organisations at large, should technology be deployed without safeguards or worker input.

Zed Law principal Nandan Subramaniam agreed that current regulations are missing the mark when it comes to the reality of AI use in the workplace.

AI, he said, “is being used wherever employees can get away with using it without having to ask”.

“In knowledge work, adoption among the people who reach for it is close to saturation. Writing, summarising, research, first-draft anything, sifting through email and documents. EY’s research has roughly half of Australian workers either not permitted to use AI at work or given no clear reason to,” he said.

Back in August 2025, EY labelled this trend a “trust crisis” whereby workers were either concerned about breaking rules or feeling largely unsupported by leadership, creating a lack of confidence and patchy levels of productivity in the workplace.

Subramaniam said: “Where it’s barely touched is anywhere that involves regulated trust. Clinical decision making, child protection, frontline advocacy.

“The real divide isn’t industry. It’s between workplaces that have given people clear permission and a sandbox to use AI well, and workplaces that have a vague ‘be careful’ memo and a lot of shadow AI nobody admits to.

“That’s where the risk actually lives.”

The risk can be mitigated by open communication, actual deference to employees, and frameworks that actually target where AI is being used by employees.

EY’s Oceania regional chief technology and innovation officer, Katherine Boiciuc, said: “The findings of our latest research reveal the importance of adopting a human-centric approach towards AI in the workplace and how this can help build trust and confidence, rather than causing concern.”

The starting point, the Curtin report explained, should be ensuring genuine consultation with the workforce, with Subramaniam adding: “The loudest voices in Australian AI policy right now belong to people who don’t use AI day to day. That’s the structural problem. You can’t sensibly regulate a tool you don’t operate.”

He continued: “Good AI policy has to be informed by the people who build the systems, the people who deploy them, the workers subject to them, and the people who get harmed when those systems fail.”

The NSW government took a significant leap in worker protection with its introduction last year of the Workplace Health and Safety (Digital Work Systems) Act, but the work is far from over.

A landmark shift in digital safety regulation, the amendment outlined that employers cannot outsource safety responsibilities to software, but drew criticism from bodies such as the Council of Small Business Organisations Australia (COSBOA), which condemned the predicted increase in compliance obligations for small businesses and their potential exposure to penalties.

Similarly, the Australian Industry Group noted that employers will be faced with increased regulatory scrutiny and union interference and disputation.

Regarding the Workplace Surveillance Act in NSW, the new digital work systems duty, and the Privacy Act amendments coming this December, Subramaniam said that much of the responsibility for protecting workers lies with employers and HR managers.

As such, he recommended making simple and easily communicable changes: “Map where AI is actually being used across the business, including the personal accounts no one is admitting to. You can’t manage what you can’t see.”

“Second, write a usable AI policy in plain English. Two pages, not thirty. Third, anywhere AI touches hiring, performance, pay, or termination, mandate a human in the loop and a documented appeals route.”

He also urged employers to audit for bias annually, not just at procurement, to refresh surveillance notices, and to train managers as well as staff.

“The legal exposure usually sits with a manager who didn’t know what tool they were using,” Subramaniam said.

In conjunction, Subramaniam acknowledged the importance of systemic change, the most important of which is federal harmonisation.

He said: “We’re heading towards eight different state and territory regimes for digital work systems sitting on top of a federal layer. National employers don’t have eight HR teams. They have one.”

“Second, accountability for algorithmic decisions has to be made genuinely enforceable. At the moment, if an AI tool sifts a candidate out of a recruitment process for a discriminatory reason, the tort and discrimination frameworks don’t fit neatly. The harm is real. The cause of action is awkward.”

He continued: “Finally, we need to be honest about who pays for compliance. If the cost of regulating AI well is borne entirely by employers, and ultimately by employees, we’ll get the worst of both worlds. Heavy compliance and bad outcomes.”

Subramaniam also acknowledged that “unions have a role. Privacy and human rights regulators have a role. So do engineers and frontline workers, and right now, they’re often missing from the table.”

As surmised by SDA national secretary Gerard Dwyer: “Technological change is not inherently good or bad; outcomes depend on how it is designed, introduced and governed, and whether workers have a genuine voice.”

Want to see more stories from trusted news sources?
Make HR Leader a preferred news source on Google.