For all the talk of how extraordinary and transformative artificial intelligence (AI) can supposedly be, it continues to be a scourge for the Fair Work Commission (FWC).
In the recent FWC case of Emmanuel Tischler v Avada Traffic Pty Ltd [2026] FWCFB 174, the Full Bench considered an appeal in an unfair dismissal matter initiated by the Applicant employee who had been unsuccessful at first instance.
In the brief published decision, the FWC set out the role AI had in the conduct of the appeal by the Applicant.
At paragraph 3 the Full Bench noted:
“Mr Tischler’s original notice of appeal contained several brief paragraphs that did not advance any coherent contention of error in the Commissioner’s decision.”
Then something changed – it seems the Applicant started using AI for the purpose of the case. As the Full Bench observed (at paragraph 3):
“Mr Tischler later filed a revised notice of appeal containing nine appeal grounds and submissions on the public interest. Mr Tischler explained that he had copied and pasted the decision into Google AI and that his revised notice of appeal ‘reflects what was found’. We will not recite this.”
So, how did that go for the Applicant? The Full Bench gave a scathing critique of the AI output (at paragraph 3):
“The nine grounds variously contend that the decision is affected by errors of law, errors of logic, and significant errors of fact. Having considered each ground, we do not find any of them to present an arguable case of appealable error. We note that the grounds contain obvious errors that in our experience are typical of AI generated applications, including the use of legal terminology in the wrong context (for example, s 387(a) does not establish a ‘threshold’), and mistaking the Commissioner’s summary of the parties’ evidence and argument as findings and conclusions made by the Commissioner.”
The Full Bench held there was no appellable error in the factual findings made by the Commissioner at first instance and, further, no public interest in even granting permission to appeal. In that regard it was noted (at paragraph 4):
“We do not consider that it is in the public interest to grant permission to appeal. Mr Tischler’s contentions about the public interest are wholly unconvincing, particularly in his (or rather, Google AI’s) assertion that the decision creates a ‘dangerous precedent’, an ‘unsafe principle’, and a ‘legally unsafe rule’. Again, the limitations of AI are apparent. The decision did none of these things. The case turns on its own facts. There is nothing to engage the public interest here, whether in connection with any of the matters referred to in Makin or otherwise.”
Accordingly, permission to appeal was refused and the matter dismissed.
The unreliability of AI was also evident in the unfair dismissal case of Sarah Byrne v CXC Corporate Services Pty Ltd [2026] FWC 2751, in which Deputy President O’Keeffe grappled with submissions that had been prepared by the Applicant employee with the (dubious) assistance of AI.
In these submissions the Applicant cited three authorities to support her contention that a modern award covered her employment as a WHS Advisor.
Deputy President O’Keeffe considered the first of these authorities, observing (at paragraph 18):
“Turning to the authorities above, I note firstly that Ponce v DJT Staff Management Pty Ltd is a case which exists. Its proper citation is Ponce v DJT Staff Management Services Pty Ltd t/as Daly’s Traffic [2010] FWA 2078. However, it is not, as the Applicant submits, a case where a WHS Advisor employed by a labour hire company was found to be award-covered. It is a case about a traffic controller who was directly employed by a company providing traffic management services. The issue to be decided was whether he was a regular and systematic casual and the issue of award coverage did not even arise. As such, it has no relevance to the award coverage issue.”
This prompted Deputy President O’Keeffe to note (at paragraph 19):
“I put to the Applicant at hearing that, contrary to her AI undertaking, she had not checked this case or indeed even read it. The Applicant provided a response that began a pattern of evasiveness displayed at regular intervals thereafter. Suffice to say, I conclude from the totality of her responses on this issue that she had not actually read the case. Having established this, I turn next to the case of “Rahman v Active Labour Pty Ltd”.”
Things went from bad to worse when the Deputy President considered this purported second “authority” (at paragraph 20):
“I scoured several legal databases to find this case without success. Having done so, I was reasonably confident that the case did not exist. My conclusion was confirmed by the admission from the Respondent’s counsel that his firm had likewise been unable to find any trace of the case. Given this, I am satisfied that this case does not exist and that the Applicant did not – and indeed could not – read it and confirm its contents.”
At least the third authority existed, albeit with a different name to that to which it was referred and not standing for the proposition for which it was cited (at paragraph 21):
“The third and final case was “CXC / Calibre Line”. I could find no such case in the usual legal databases. In discussing this case with the Applicant – who claimed to have seen it – it emerged that the case was actually Mr Allan Gabriel v Titan Recruitment Pty Ltd T/A Titan Recruitment [2023] FWC 230. The term “Calibre” came from a company that was somewhat involved in the matter. Having been directed to this case – and remembering it was cited in favour of WHS roles not being managerial in the context of award coverage – I was somewhat bemused to find that the actual case involved considering an extension of time for a s.365 general protections application. Once again, the case had no relevance whatsoever to the matter at hand and the Applicant had clearly not read it to any meaningful extent.”
Deputy President O’Keeffe rightly excoriated the Applicant for citing these three “authorities” (at paragraph 22):
“In summary and notwithstanding her AI undertaking, the Applicant presented three cases in support of her argument. One simply did not exist. One existed under another name but did not have the slightest relevance to this matter. The third existed and the name was correct but again, there was no relevance to the present case. Somewhat gallingly, in the face of these revelations the Applicant appeared unwilling to admit that she had not actually read these cases to check them, as she had undertaken to have done. She tried to prevaricate and suggested that perhaps she had only read parts of them or had simply read summaries.”
Deputy President O’Keeffe then continued his criticisms (at paragraph 23):
“I make it clear that I do not accept the Applicant’s answers. It is simply not credible to suggest she had ready any of the cases or their summaries and then made the claims about their content she made in her submissions. As such, the value of the Applicant’s submissions with respect to case precedent for her being award-covered is nil. I further note that in submitting these irrelevant and non-existent cases the Applicant has wasted significant time both for the FWC and the Respondent’s counsel.”
Some Observations
These cases are not merely isolated examples of the detrimental effect the use of AI has had on the conduct of cases before the FWC. They are reflective of a general pattern. No doubt there will be more such cases as the use of AI in FWC proceedings continues, seemingly unabated.
Costs orders are, for good policy reasons, rarely made in unfair dismissal matters. The FWC should, however, have no compunction about exercising the power it has to award costs in unfair dismissal proceedings (including under section 400A of the Fair Work Act for an “unreasonable act”) when parties use AI to produce submissions that are devoid of merit or, worse still, cite authorities that either don’t exist or clearly do not support the proposition for which they are being relied upon. It wastes the time and resources of the FWC and the hapless employers who need to respond to the AI slop served up to them. To compound the problem, submissions produced by AI are not only often devoid of merit but can also be voluminous and largely incomprehensible. The uncritical use of AI by applicant employees is a scourge for the FWC and employers. It has been fuelled by spurious claims from vested interests about what AI is presently capable of that are so hyperbolic they would make the snake oil salesmen of old blush. This is not the time for timidity — in circumstances where the gross misuse of AI is evident in unfair dismissal proceedings employers should make applications for costs and the FWC should order costs be paid. Some adverse costs orders, well-advertised by the FWC, might have a salutary effect on the tide of AI-driven applications currently overwhelming the tribunal.
