Fair Work Commission condemns 'plain wrong' AI legal advice

I found this formal decision PDF for the Khan case: https://fwcgovau.sharepoint.com/teams/group252c29c/_layouts/...

Key section:

> Two days after Mr Khan filed his claim, Fair Work Commission staff sent Mr Khan an email advising that the Commission may not have the power to deal with his case because he had not worked for the minimum employment period. [...]

> In the morning of Friday 31 July 2026 I took the time to write to Mr Khan and explain to him in plain English why his case was unlikely to succeed. Mr Khan was encouraged to discontinue his application, and advised that there could be costs consequences if he did not discontinue his claim. [...]

> Khan sent a reply email that same morning. Mr Khan’s AI generated email once again addressed the wrong end date and repeated the same irrelevant AI-generated arguments. Mr Khan’s email did not engage, at all, with the matters raised in the earlier email from the Commission

> A sacked ALDI employee condemned by the Fair Work Commission for using "plain wrong" artificial intelligence (AI) advice in his failed bid to reverse his dismissal says he now plans to use "two or three" different AI agents to help him appeal against the decision.

> Sadnan Khan was ordered to pay $1,230 towards the supermarket giant's legal fees after a landmark Fair Work Commission decision found he used AI as a "quasi-legal advisor" to challenge his dismissal, despite repeat warnings his case had "no substantial prospects of success".

> "I have taken the very rare step of ordering him to pay some of the employer's legal costs, because his unreasonable conduct caused his former employer to incur those costs unnecessarily," Fair Work Commission deputy president Michael Easton said in the tribunal's decision last week.

The article is balanced; it also cites the case of a computer science lecturer at Macquarie University who successfully challenged the university's treatment of him using AI. (Over the past decade or so, Mac Uni has become very attached to the idea that minimum wage is enough for academic staff.)

So it's the same story as we've been hearing for more than a year now: LLMs are only useful if you ignore all the fluff where they tell you how great everything is, and know how to extract and interpret the parts where they say something of substance, and actually do the work of determining whether the thing is bullshitting you.

> Research commissioned by the workplace umpire shows its cases have recently grown by 40 per cent, in part due to litigants using AI.

A recent thought I had about the AI rollout: it’s such a massively unbalanced bet. We are experiencing right now the negative aspects of AI, we are paying for that. For a chance that maybe some hypothetical, sci-fi-like technology exists in the future. So we know we will always suffer the losing end of the bet, and we have no idea if we will win, or even how that would look like to win.

In this example we see how the commission is under stress, something we see across the board, and the benefits so far have been that software and emails are written faster, and maybe something more will come at some point in the future (that we also know will have its own set of externalities).

It’s insane that such a small group can decide to play that gamble for the rest of society, worldwide

> The tribunal acknowledged AI was improving access to justice for applicants with genuine, meritorious claims "who might otherwise have been deterred".

It's good the tribunal acknowledges we can't all necessarily afford the costs you may incur in those initial discussions with a lawyer. While there are pro-bono lawyers they have to make sure they're taking the cases with the best prospects for success so they don't go out of business themselves.

> "Most community members are aware of this kind of property of AI now where it has a kind of sycophantic tendency … and may in fact encourage people to believe that there are legs in a claim they're making when that may not in fact be the case."

It'll be interesting to see how the AI providers are going to balance the need for models to be engaging enough to keep the tokens flowing but not so engaging they give bad advice to encourage continued conversations.

The second case mentioned in the article is interesting because no other case about employee choice had gone to merits before.

Thus it might be the first example of AI winning a case that sets some kind of precedent.

At the very least it shows that AI can put together coherent legal arguments even when there are no close precedents or case law it can refer to. This was something a lot of lawyers didn't believe would be possible for AI to do.

> applicants to the Fair Work Commission will need to disclose AI use

I really don't like where this is going.