Compliance-heavy tasks like contracts and award interpretation still need people, say HR experts
Employers who rely on generative artificial intelligence (AI) to draft employment contracts or interpret awards are taking on risk they cannot hand back to the machine, HR leaders on both sides of the Tasman warn.
Australia’s workplace tribunal now takes the same view. New Fair Work Commission rules on the use of generative artificial intelligence (GenAI) took effect on 20 October 2026, placing fresh disclosure and verification obligations on employers, employees and the HR teams that represent them.
The pressure on the Commission is significant. AI-assisted Fair Work applications are surging, with signs of AI use reported in 50–75% of matters and lodgements rising by more than 40% in three years, which the Fair Work Commission has linked closely to the emergence of generative AI.
For HR leaders, the question is no longer whether to use AI, but where human judgement has to take over.
Where should AI stop and human judgement start?
Mollie Eckersley, associate director of operations, ANZ, at BrightHR Australia, says the dividing line comes down to risk.
“I think it’s best used on tasks that are generally low risk. So your admin tasks that are also reversible,” she said, pointing to summarising lengthy information, structuring meeting notes and drafting first versions of company-wide communications.
“Where I don’t think we should be relying on AI [is] when it comes to the complexities of drafting employment contracts, drafting policies, drafting procedures, award interpretation, pay rates. That’s where it gets really, really complex.”
Responsibility for getting those things right does not shift to the software, Eckersley said. “It is on the employer to make sure they’re determining and paying their employees correctly, making sure they’ve got the right contracts, the right policies in place.”
Why small businesses face greater compliance risk
BrightHR runs an employment relations advice line and Eckersley said a familiar pattern has emerged. Family-run businesses grow, then seek help after relying on online templates or tools such as ChatGPT.
“There’s cases that have gone through to Fair Work where you can see AI has clearly gotten it very wrong,” she said. “Missing information, missing context, maybe interpreting legislation incorrectly.”
Smaller employers are more exposed because they lack in-house expertise. As HRD reported when the Fair Work Commission AI disclosure rules came into force for employers, the obligations apply to employers defending claims as much as to workers bringing them.
What good human oversight looks like
Eckersley says effective oversight is an active process, not a rubber stamp. “It’s more than just someone clicking approve and saying, yep, no problem, that contract’s good to go,” she said.
“I think the human touch and the conversations there unravel a lot more details regarding an employee and their circumstances than you ever would get through AI.”
Policies also need regular review. The Guidance Note itself acknowledges that, given the pace of GenAI’s evolution, these obligations will evolve too.
Building AI guardrails employees will actually use
At Wellington-based digital agency Alphero, the challenge is as much cultural as technical. Lisa Phillips, head of people and culture (Kaitaki Waka), said junior and intermediate developers fear that working with AI will eventually cost them their jobs.
Her message to them: “We’re still going to need increased amounts of quality engineering, increased amounts of architecture being done well.”
Alphero has rewritten every job description this year. Roles that were once defined by craft now give equal weight to mindset and behaviour. “Curiosity, inventiveness, bravery, all of those things are now equally weighted in our JDs alongside being technically good because those human things are just becoming more and more important,” Phillips said.
The agency’s operational AI policy came out of cross-functional workshops with its heads of security and architecture. Client demands drove much of the detail. “A heap of our clients are like, don’t put any of our stuff anywhere near AI,” she said. “I think we had probably 11 or 12 versions before everyone was happy with it.” Each draft was stress tested with non-technical staff to make sure it made sense.
The result is a traffic light system that the security team keeps up to date. “You can use green light company license for all of these tasks, orange light check with security for any of these tasks,” Phillips said. A separate section flags clients whose data must not go into any AI tool. Alphero is also building tooling so teams can see their token usage, after cautionary tales of runaway costs.
Experimentation is encouraged within those limits. Staff share AI projects internally and present at regular show-and-tell sessions on how they are integrating AI into their work.
That pairing of clear rules with visible experimentation addresses a gap that HRD has covered before, including AI policy gaps putting New Zealand businesses at risk and research suggesting AI governance is maturing faster than employee guidance.
For Eckersley, the bottom line for employers of any size is the same: “It’s really important to have that support and have that backing and not just put all of your trust into AI.”
