Judge issues “cautionary tale about the dangers of befriending Artificial Intelligence (AI)-powered chatbots who masquerade as legal advisors”

Background

The Applicant, Mr Ba, commenced proceedings in the Federal Circuit and Family Court of Australia against Sterling Parts Australia Pty Ltd and Sterling Parts (Melbourne) Pty Ltd. Mr Ba had performed work for the respondent entities as a delivery driver. His claims broadly included a claim that he had been dismissed in contravention of the general protections provisions of the Fair Work Act 2009 (Cth) and that he had been incorrectly classified as an independent contractor, when instead he alleged he was an employee of either of the First or Second Respondents. He also claimed alleged underpayments due to classification errors, as well as unpaid annual leave and superannuation.

In support of his claim, Mr Ba relied upon a Schedule of Documents (Evidence List) containing 38 separate entries that each referred to a document that was said to prove a “Key Point” in his claim. Each document in the Evidence List was assigned a reference number. Those reference numbers appeared throughout Mr Ba’s Statement of Claim.

The Respondent entities made various efforts to attempt to obtain the documents referred to in Mr Ba’s Evidence List, including written requests to Mr Ba, and serving a Notice to Produce.  The Notice to Produce again required Mr Ba to produce the documents referred to in his Evidence List.

In one of the written requests, the Second Respondent’s solicitor (Mr Tang) noted to Mr Ba that he had previously referred to his artificial intelligence (AI) chatbot as his “industry expert” in correspondence with Sterling Parts before the proceedings commenced. Mr Tang also referred Mr Ba to the Court’s Practice Note concerning the responsible use of AI. Mr Tang cautioned Mr Ba that continued misuse of chatbots in the proceeding, including by having it generate evidence, might result in an application for costs being made against him.

Application to Dismiss

The Respondent entities ultimately sought orders from the Court requiring Mr Ba to produce each document identified in the Notice to Produce, and if Mr Ba failed to comply, that Mr Ba’s claim be dismissed. After hearing from the parties, the Court made Orders to this effect.

After the Court made those Orders, Mr Ba produced a “Submission Letter”. The Submission Letter referred to photographs and an email which was sent by the Second Respondent. Mr Ba did not produce the photographs he referenced, and other than the email, again failed to produce any other documents from the Evidence List.

In light of Mr Ba’s non-compliance, the Court determined to dismiss Mr Ba’s claim.

AI is not a legal advisor

In its judgment, the Court described Mr Ba’s conduct as demonstrating a “pattern of obstruction and prevarication” and stated it had “little hesitation” in granting the Respondents’ application to dismiss his claim. The Court also noted that Mr Ba’s continued reliance upon AI-generated documents that could not be produced or verified had “arguably gave rise to an abuse of process”, which came perilously close to involving a contempt of Court (and the penal sanctions which can potentially apply).

The Court acknowledged that generative AI has the potential to improve efficiency and access to justice, particularly for self-represented litigants and those for whom English is not their native language. However, the decision reinforces that responsibility of all litigants for ensuring the accuracy of documents such as pleadings, submissions and evidence, and the significant consequences which can follow from blind reliance on AI in running a claim.

Importantly, the Court also made it clear that AI is not a legal practitioner. Unlike legal practitioners, AI does not owe any overarching duty to the Court or to the proper administration of justice, does not hold a practising certificate and is not authorised to provide legal advice in any Australian jurisdiction. AI is also not subject to the professional and ethical obligations that govern the legal profession. By contrast, the Court referred to the “sycophantic tendency” of AI large language models to prioritise user approval over truth, which was said to make them “especially dangerous and unpredictable unless careful oversight, calibration and verification is applied”.

Both Respondents denied that Mr Ba’s claim had any legal merit. However, Mr Ba did not have the benefit of the Court’s adjudication of his claim as a result of his use of AI in preparing his claim and referring to documents which the Court inferred did not exist (save for the one document he produced). The Court emphasized that it would be an unsatisfactory outcome for litigants if the misuse of AI resulted in good (or at least arguable) claims being “thrown on the proverbial scrap heap”.

Key takeaways

The Court’s decision in Ba v Sterling Parts Australia Pty Ltd provides one of the clearest judicial warnings to date about the risks associated with the use of generative AI by unrepresented litigants.

There is no doubt that the use of artificial intelligence can be a very effective tool, particularly when it is used by legal practitioners who are familiar with the substantive law, and the practice and procedure of the jurisdiction in question. However, use of artificial intelligence by self-represented litigants can pose particular challenges for employers in responding to claims. Employers should:

• Ensure that they are aware of and comply with any Practice Notes or other restrictions which inform the manner in which AI can be used in the relevant court or tribunal.
• Ensure they scrutinise all cases and legislation cited by the employee carefully. It is well known that AI can “hallucinate”, including by fabricating case citations or referring to legislation that does not exist. Employers should ensure that the cases and legislation cited by an employee in fact exist, support the proposition made, and have been accurately characterised.
• Ensure they verify that any documents referred to by the employee in their pleadings and evidence exist and stand for the propositions made by the employee. If a respondent employer does not have copies of documents referred to by an employee in their pleadings, copies of those documents should usually be sought ahead of the filing of a defence (in court proceedings).
• Focus on responding to the underlying claim, rather than on the suspicion that AI was used to prepare it. Particularly in proceedings before the Fair Work Commission, the substantive allegations should be assessed independently of the quality of the drafting.
• Where appropriate, seek that the court or tribunal holds the employee to account. For example, if the employee has failed to comply with the requirements of that jurisdiction concerning use of AI, or where they have referred to legislation, case law or documents which do not exist.
• Expect that additional time and resourcing will be required to respond to voluminous AI generated claims, and plan ahead. For example, it may be reasonable to seek longer timeframes for the filing of a defence and evidence, having regard to the nature of the material filed by an employee.

If you have any queries about this decision for your business, please do not hesitate to contact a member of Piper Alderman’s Employment Relations Team.