The Fair Work Commission has warned against treating AI as a shortcut in workplace matters.

Its draft guidance says AI can be helpful, but it is not a substitute for legal advice, award interpretation or HR compliance expertise, and that AI-generated material may be “inaccurate, incomplete, out of date or just made up.”

The biggest danger with using AI in these scenarios is that it is plausible advice that sounds polished, confident and legal, but is still wrong in ways that matter. The Commission has also warned that AI tools may give unrealistically optimistic predictions about prospects and compensation, which can encourage claims that should not be pursued. This is not a fringe issue either. The Commission’s own workload has increased by over 70 per cent in three years, a rise its President has linked directly to the growing use of AI tools by people preparing claims (Source: Fair Work Commission).

Where AI gets it wrong

Employment issues are rarely straightforward. Before anyone drafts a letter or starts a process, someone needs to identify the real issue, the right law and the right risk.

AI often gets this wrong by:

  • citing the wrong section of the Fair Work Act or applying a section that does not fit the facts;
  • relying on outdated law in fast-moving areas such as casual employment, fixed-term contracts, flexible work, labour hire and general protections;
  • applying the wrong jurisdiction, including overseas or state-based principles that do not fit the national workplace relations system;
  • inventing or misquoting cases, or citing decisions that do not support the point being made.

It can also miss the threshold question entirely. Is the employee in the national system? Does a modern award or enterprise agreement apply? Have they completed the minimum employment period? Is the issue really unfair dismissal, or is it general protections, discrimination, workers compensation or adverse action? If the foundation is wrong, the answer built on top of it is wrong too.

This is not a rare glitch. Stanford University research found that general-purpose AI tools hallucinate on legal questions between 69 and 88 per cent of the time, inventing authorities or misstating the law. In an area as technical as employment, that is not a margin of error you can afford to build decisions on.

Awards are not simple

Modern awards are highly technical. Coverage depends on the employer’s industry, the employee’s actual duties, classification definitions, exclusions, higher duties, allowances, overtime triggers, rostering provisions and the interaction with the National Employment Standards.

That is not a keyword exercise. It requires legal and practical analysis. Even experienced HR professionals need time to work through award coverage properly, and AI often skips over the nuance that decides the outcome. Even a tool built specifically for the task cannot be trusted to get this right on its own.

Why employers should care

A wrong AI answer can create real exposure. It can lead to underpayments, incorrect classifications, invalid contract terms, flawed termination processes, sham redundancy risks, missed consultation obligations or unlawful deductions.

Dismissal matters are especially risky. AI may produce a professional-looking letter but miss award obligations, procedural fairness, redeployment issues, the employee’s history, or whether the employee has exercised a workplace right. That is how a tidy draft turns into a messy claim.

Contracts and policies are no safer just because they read well. AI can produce clauses that are unenforceable, inconsistent with the National Employment Standards, inconsistent with an award, or too generic to reflect the business’s real processes.

A live example

In Hoverd v M & J D Pty Ltd FWC 1013, the applicant admitted using AI to draft submissions and relied on contract clauses and award provisions that did not exist. The Commission dismissed the application and invited the employer to seek costs.

That matters because it shows how easily AI can generate material that sounds legal but is disconnected from the real contract, award or facts. If that can happen in Commission submissions, it can happen just as easily in an internal HR process.

Privacy matters too

There is also a confidentiality problem. The Fair Work Commission’s draft guidance warns users not to provide personal or confidential information to public AI tools.

That is especially relevant in HR, where matters often involve medical information, payroll data, complaint material, disciplinary history and sensitive investigations.

If it should not be shared broadly, it should not be going into a public AI prompt.

Closing

AI can be a useful assistant in various aspects of business, but it is not an HR adviser, employment lawyer or award interpreter. In Fair Work matters, the cost of a confident wrong answer can be far greater than the time saved. And in the time you have spent prompting it over and over, you probably could have picked up the phone and called us.

This article provides general information only and does not constitute legal or professional advice. Before making any decisions in a workplace matter, always consult a qualified HR or workplace relations professional, and where appropriate seek legal advice. For guidance tailored to your business, get in touch with the team at Inject.