The AI Illusion: Why Self-Represented Litigants Are Losing in Court

Can a chatbot really outperform your lawyer? Self-represented litigants seem to think so.

Workplace claims filed by self-represented litigants have more than doubled in the Federal Court over the past year, driven by the compelling shortcuts delivered by artificial intelligence (AI).

The Hallucination Problem

AI's most dangerous flaw in litigation is its tendency to fabricate. AI produces "hallucinations", unsourced and incorrect information generated by limitations in the data used for its training, including entirely non-existent case law. This is not a theoretical risk. In Howe v Secretary, NSW Department of Education, a self-represented applicant filed submissions containing fictionalised case references that the respondent could not locate, because they did not exist¹.  The respondent observed that the applicant had "utilised generative AI to prepare" his submissions but "not taken any care to ensure that the submissions he is advancing to the Commission are based on actual legal authority or principle."²

In Sharma v SafeWork NSW, another self-represented litigant filed AI-generated submissions and appeared "not to understand the requirements of Practice Note 33" ³.  In Shahin v Industrial Relations Secretary on behalf of Multicultural NSW, an applicant's "unchecked use of generative artificial intelligence" rendered evidence inconsistent and unreliable⁴.  In Sylvester v Health Secretary in respect of HealthShare NSW, the tribunal rejected an applicant's attempt to rely upon new evidence, in part because of "the use of generative AI in its preparation" ⁵. A troubling pattern is emerging across the Courts: self-represented litigants continue to fall for AI’s illusion of legal competence, trusting its authoritative tone without recognising the fabrications beneath.

Courts are Watching and the Rules Are Tightening

Practice Note 33 of the Industrial Relations Commission of NSW now requires any party using generative AI to verify all citations and legal references, and that verification "must not be solely carried out by using a Gen AI tool or program" ^6,This obligation applies equally to self-represented parties. The Federal Court's Practice Note GPN-AI, issued in April 2026, imposes its own requirements around source traceability and human supervision.⁷

What AI Cannot Do

Filing a claim is the easy part. Winning one requires something AI fundamentally lacks, being professional judgment, strategic thinking, and the human understanding that litigation demands.

AI lacks the ability to exercise the higher judgment and ethical obligations required for legal services. It cannot build the trust and rapport that are central to the solicitor-client relationship, nor can it replicate the emotional intelligence needed to navigate sensitive disputes. In situations of legal ambiguity, AI frequently provides incorrect responses and, unlike a lawyer, does not seek further context to ensure the best possible outcome.

In Rizkalla v CDC Geelong Pty Ltd [2026] FCA 249, a Judicial Registrar described a self-represented applicant's concise statement as "poorly formulated" and "strongly recommended that the applicant obtain legal advice."  The Registrar distilled the applicant's entire claim into a few sentences of guidance, being something a solicitor would have done before the first document was filed.

Under section 596(2) of the Fair Work Act 2009 (Cth), the Fair Work Commission may grant permission for legal representation where complexity demands it, where a person cannot represent themselves effectively, or where fairness requires it. These provisions recognise a fundamental truth: the law is not a product to be assembled from prompts, it is a discipline requiring skilled human application.

The Real Cost of Going It Alone

AI may lower the barrier to filing a workplace claim, but it raises the stakes dramatically for those who proceed without professional guidance. A fabricated citation can destroy credibility. A flawed cause of action could be fatal to a case. Failing to comply with a practice note can see evidence struck out before a hearing begins.

No chatbot bears the consequences, that burden falls entirely on the litigant!

Solicitors and barristers remain bound by strict professional and ethical obligations to act in their client's best interests, obligations that AI owes to no one. In litigation, the value of a skilled legal professional has never been clearer, and the cost of doing without one has never been higher.

A Question Worth Asking

The uncomfortable irony at the heart of the AI litigation boom is that the very technology that promises to democratise access to justice may be doing the opposite. Every poorly drafted claim clogs court lists, delays genuine disputes, and invites costs orders against litigants who may not be able to afford same.

Every fabricated citation erodes judicial trust in the submissions that follow, and every dismissed case is a person who believed they had rights worth enforcing, and who may well have been correct, but who will never know, because their case was lost on process before it was ever heard on merit.

The question is not whether AI has a place in the legal system. It does. The question is whether we are prepared to let people mistake a shortcut for a strategy and pay the price when the courtroom reminds them that the law has never been as simple as a prompt and a response.

‍

‍

‍

1. Howe v Secretary, New South Wales Department Of Education [2025] NSWIRComm 1081, [16].
2. Ibid [17]. Sharma v SafeWork NSW [2026] NSWIRComm 1056 [14].
3. Shahin v Industrial Relations Secretary on behalf of Multicultural NSW [2026] NSWIRComm 1021, [41].
4. Sylvester v Health Secretary in respect of HealthShare NSW [2026] NSWIRComm 1013, [28].
5. Industrial Relations Commission of New South Wales, Practice Note 33, [19].
6. Federal Court of Australia, Use of Generative Artificial Intelligence Practice Note (GPN-AI).
7. Rizkalla v CDC Geelong Pty Ltd [2026] FCA 249, [18].