Judge says reliance on misleading evidence list came close to contempt of court, abuse of process
In a proceeding where the Federal Circuit and Family Court of Australia’s second division dismissed an unlawful termination claim, the judge described the case as “a cautionary tale about the dangers of befriending Artificial Intelligence (AI)-powered chatbots who masquerade as legal advisors.”
In Ba v Sterling Parts Australia Pty Ltd [2026] FedCFamC2G 1245, the applicant alleged that he was an employee, specifically a delivery driver, of either the first or the second respondent, which controlled his work via an integration national system.
According to the applicant, the respondent terminated his work by denying him access necessary to perform his duties after he raised concerns and made inquiries about his working arrangements, conditions, classification as an independent contractor, and workplace law entitlements.
The applicant contended that the respondent incorrectly considered him an independent contractor and dismissed him in breach of s 340(1) of the Fair Work Act 2009 (Cth). To support his claim, he filed four documents, including an evidence list, with the court on 2 February 2026.
In denying the claim, the Federal Circuit and Family Court of Australia (Division 2) determined that the applicant demonstrated a pattern of obstruction and prevarication that began on 30 March 2026 and included his failure to obey a 17 June 2026 order.
Under the 17 June 2026 order, the applicant had to produce copies of the documents included in the evidence list he had submitted or to explain whether any of them were beyond his control. The court described his suggestion that he did not understand what he needed to provide as disingenuous.
The court then addressed the applicant’s apparent reliance on generative artificial intelligence (genAI). The court accepted the benefits of responsibly using genAI tools in terms of efficiency and access to justice.
“However, there are significant difficulties when AI answers a job description for legal advisor,” wrote Judge Catherine Symons for the court. “This is because AI is not a lawyer, does not hold a practising certificate and is not authorised in any jurisdiction in Australia to engage in legal practice and to provide legal advice.”
The court stressed that AI did not care about the truth of generated material and owed no duty to the court or the administration of justice to act in a litigant’s best interests or to meet regulatory or ethical standards.
The court emphasised that large language models tended to prioritise user approval over the truth, which rendered them dangerous and unpredictable without careful oversight, calibration, and verification.
The court saw no evidence that the applicant had taken steps in accordance with the Federal Circuit and Family Court of Australia’s practice direction on AI use, issued on 29 May 2026, to assess:
The court noted that the applicant failed to produce any documents except for the second respondent’s termination email and doubled down on the issue of documentary production.
Thus, the court inferred that an AI tool utilised by the applicant generated the evidence list, which referred to unidentifiable documents that did not exist and/or did not bear the assigned description. The court ruled that the reliance on this misleading evidence list:
Given the applicant’s choice to put blind faith in his AI “legal expert,” the court declined to adjudicate the respondents’ assertions that the applicant’s claim lacked any merit.
“While AI has the capacity to flatter and beguile, it also has the capacity to obscure and undermine,” said Judge Symons. “It would be an unsatisfactory outcome indeed if the misuse of AI resulted in a good (or at least arguable) claim being thrown on the proverbial scrap heap.”
