The cost of getting AI wrong: recent decisions and new rules for Fair Work Commission proceedings

The Fair Work Commission (FWC) is facing a deluge of applications. Its total workload increased by more than 70% in the three years to June 2026, a rise that Commission President Justice Hatcher attributes principally to the growing use of generative artificial intelligence (AI). Likewise, the Federal Court has experienced a significant increase in self-represented employment litigation. Between 2024 and 2025, its Employment and Industrial Relations National Practice Area recorded a 200% increase in matters allocated to judges involving self-represented applicants.

The proliferation of AI also has implications for the conduct and cost of proceedings. Recent FWC decisions have identified reliance on fabricated or mischaracterised legal authorities, non-existent contractual and award provisions, and potentially AI-assisted correspondence that failed to address the substantive issues in dispute. While AI can improve access to justice, the FWC's research found that uncritical reliance on AI-generated material can contribute to unrealistic expectations, claims with weaker prospects of success and more complex case management. These issues place additional pressure on the FWC's resources and may increase the costs incurred by employers responding to inaccurate or irrelevant material.

To address these issues, the FWC recently introduced the Guidance Note: Use of generative artificial intelligence in Commission cases (Guidance Note). The Guidance Note requires parties using AI to prepare documents for FWC proceedings to:

• disclose AI use, including how it assisted in preparing the document;
• independently verify the accuracy and relevance of their material, including cited authorities, factual assertions and supporting evidence, and acknowledge completion of these verifications; and
• verify witness evidence, ensuring AI-assisted statements reflect the witness's own knowledge and words.

The guidance follows several decisions examining the consequences of inappropriate AI use, which resulted in adverse costs orders against applicants who commenced claims with no reasonable prospect of success or who unreasonably continued proceedings despite clear warnings about their prospects.

When can the FWC award costs?

Parties to FWC proceedings ordinarily bear their own legal costs. However, the FWC may award costs where an application is made:

• vexatiously, primarily to harass, embarrass or obtain a collateral advantage;
• without reasonable cause, having no substantial prospect of success when commenced;
• without reasonable prospect of success, where it should have been reasonably apparent to the applicant that their claim was manifestly untenable or groundless.

Additionally, in unfair dismissal proceedings the FWC may award costs where a party's unreasonable act or omission in conducting or continuing proceedings causes another party to incur costs.

Pursuing claims with no reasonable prospect of success

In Singha v Metal Manufactures Pty Limited [2025] FWC 2817, the FWC awarded costs against an applicant who pursued an unfair dismissal claim despite being ineligible because his earnings exceeded the high-income threshold and he was not covered by an award or enterprise agreement.

Despite express warnings from the FWC about the jurisdictional issue and potential costs consequences, the applicant continued to pursue his claim, largely on grounds non-responsive to the threshold question.

Following dismissal of the claim, the employer applied for costs. In the costs proceedings, the applicant acknowledged using AI to prepare his submissions, which contained several deficiencies, including:

• references to cases that did not exist;
• incomplete citations to legislation and case law; and
• reliance on authorities that did not support the legal propositions asserted.

Deputy President Boyce ordered the applicant to pay $7,869 in costs, finding that:

• it should have been reasonably apparent from the outset that the application had no reasonable prospect of success; and
• continuing proceedings despite the FWC's warnings constituted unreasonable conduct on the applicant's part that caused the employer to incur costs.

Similarly, in Khan v Aldi Pty Ltd [2026] FWC 3144, the FWC awarded costs against an applicant who continued an unfair dismissal claim despite not meeting the minimum employment period to bring a claim.

The applicant relied extensively on AI-generated submissions, which repeatedly addressed the wrong legal question: when his dismissal took effect, rather than when he was notified of it. Despite the FWC explaining the correct statutory test, warning of potential costs consequences and inviting him to discontinue, the applicant continued to advance irrelevant AI-generated arguments.

Deputy President Easton found that continuing the claim after receiving those warnings was unreasonable and caused ALDI to incur unnecessary costs preparing for and attending the hearing. The applicant was ordered to pay $1,230 in costs, limited to those incurred after he was expressly warned of the deficiencies in his claim.

Importantly, Deputy President Easton acknowledged that using AI to prepare submissions was not inherently unreasonable. Rather, the unreasonable action of the applicant was not taking adequate steps to critically assess the material he submitted, particularly when confronted with clear information demonstrating that his claim could not succeed.

The Federal Court's decision in Raghib v Stantec Australia (Costs) [2026] FCA 1415 further illustrates the consequences of using unreliable AI-generated material to pursue unmeritorious proceedings.

In proceedings seeking judicial review of FWC decisions, the Federal Court identified significant deficiencies in the applicant's materials, including:

• incoherent legal arguments; and
• inconsistencies between documents filed and submissions.

Although the applicant maintained that he had used AI only to check spelling and grammar, Justice Wheelahan considered it likely that some anomalies resulted from the indiscriminate use of AI.

The Federal Court found that the proceedings had been instituted without reasonable cause, as the grounds of review were fundamentally defective and bound to fail. The applicant's unreasonable conduct of the proceedings also caused the employer to incur additional costs.

The applicant was ordered to pay $45,000 in costs. In so doing, the Federal Court emphasised that self-represented litigants are not immune from adverse costs orders merely because they lack legal expertise.

Consequences beyond costs orders

The consequences of inappropriate AI use are not limited to adverse costs orders.

In Application by De Bruyn [2026] FWC 2887, the FWC found that the applicant had been subjected to workplace bullying, but also raised concerns about correspondence prepared by her husband on her behalf.

Commissioner McKinnon considered the correspondence to have been prepared using AI and that its excessive length, repetition, and accusatory tone had contributed to a breakdown in communication between the parties. Consequently, legitimate workplace concerns had gone unaddressed.

In making its stop-bullying orders, the FWC expressly prohibited both parties from using AI to prepare future correspondence between each other. The orders also required communications to remain accurate and respectful, with parties encouraged to address issues concisely and substantively.

How does the Guidance Note address issues associated with AI-generated claims?

The Guidance Note responds directly to several deficiencies identified in recent decisions. Its disclosure and verification requirements seek to ensure parties critically assess AI-generated material before filing it, rather than leaving the FWC and opposing parties to identify and address inaccuracies.

These requirements may help address:

• fabricated authorities: the requirement to verify that cases exist and support the propositions advanced directly addresses the issues identified in Singha.
• incorrect legal analysis: mandatory human review may help identify fundamental legal errors, such as those repeatedly overlooked in Khan.
• incoherent submissions: verification of accuracy and relevance may reduce inconsistent and unsupported material, as seen in Raghib.

The requirements are not merely procedural. Failure to comply, including by failing to disclose AI use, may result in documents being disregarded or given reduced weight. Depending on the circumstances, parties may also face adverse costs orders or dismissal of proceedings under the FWC's existing statutory powers.

However, compliance with the Guidance Note does not establish that a claim has reasonable prospects of success, nor does it alter the statutory grounds for awarding costs.

Ultimately, the Guidance Note may reduce the burden of inaccurate AI-generated material, but its effectiveness will depend on parties meaningfully scrutinising that material rather than treating verification as a procedural formality.

For employers handling increased volume of AI-assisted complaints, a central focus should be on the complaints handling process - with the aim of identifying issues early, before they escalate into formal legal proceedings. This is set out in more detail in our previous insight.