When AI makes it easier to litigate, who pays the price?

They need to work out whether the law provides a remedy, where to pursue it, how to articulate the claim, and what evidence supports it. For many, there is another hurdle: obtaining legal advice about whether the case is worth pursuing at all.

Generative artificial intelligence is changing that equation.

An unrepresented litigant can now use readily available technology to research potential causes of action and legal principles, organise evidence and produce pleadings, submissions, and correspondence that can resemble professionally prepared material.

What the technology cannot necessarily do is reliably tell them whether they should bring the claim in the first place.

That tension is increasingly significant for the Fair Work Commission (FWC), which received a record 53,617 lodgments in the 2025–26 financial year, up from 44,075 the year before.

FWC president Justice Adam Hatcher has said it can reasonably be inferred that increasing use of generative AI by potential litigants has principally driven the commission’s extraordinary workload increase over the past three years.

VIEW ALL

Yet Justice Hatcher has also described the technology’s ability to alert an aggrieved employee to a possible remedy and the deadline for pursuing it as a win for access to justice.

AI can help people who cannot afford legal representation understand and pursue legitimate rights. It can also allow weak or misconceived claims to reach a tribunal or court without scrutiny they may once have encountered beforehand.

The rise of the “AI advocate” has already raised questions about what growing numbers of rights-aware, self-represented litigants mean for courts, tribunals, and employers.

Increasingly, the justice system is having to contend with both consequences at once.

The disappearing filter

Kingston Reid partner Michael Stutley has seen the change not only in the number of self-represented employee litigants, but in how their cases are prepared and how far they progress.

Before the widespread adoption of generative AI, Stutley said employees were generally more hesitant to represent themselves and, where they did, applications tended to reflect their own understanding of the dispute.

“Nowadays, the use of generative AI appears to be encouraging some individuals to pursue claims they might otherwise not have brought,” he said.

“We are also seeing an increase in lengthy materials that, while sophisticated or ‘polished’ on the surface, do not always have the legal substance or relevance that the drafting style might suggest.”

Stutley said his practice had also seen a significant increase in matters proceeding to final hearing rather than resolving earlier through negotiated outcomes.

Hall & Wilcox partner Fay Calderone similarly said the shift was “not simply an increase in the number of claims, but a change in the nature of those claims”.

AI allows applicants to generate lengthy pleadings, witness statements, submissions, and correspondence that can appear legally persuasive at first glance, she said, while the underlying analysis may be inaccurate, disconnected from the facts or based on principles that do not apply.

“A matter that may previously have been resolved relatively quickly can now involve hundreds of pages of AI-generated material, increasing complexity, time and cost,” Calderone said.

Dentons partner Paul O’Halloran has seen another consequence: employment claims becoming more difficult to settle.

“I have also found that it’s much more difficult to settle employment claims, because individuals have confirmation bias from what they are being fed back to them from AI about the merits of their claims,” he said.

For Calderone, part of the problem is the absence of the “gatekeeping and reality-testing role” lawyers have traditionally performed by assessing evidence, identifying weaknesses, managing expectations and advising clients when a claim lacks merit or is not commercially sensible to pursue.

O’Halloran put it more starkly, noting that “AI has lowered the barrier to entry for all claimants, regardless of the merits of their case”.

“Claims that a plaintiff solicitor would previously have advised against pursuing are now being filed, as there is no professional filter,” he said.

Who was the filter keeping out?

National Justice Project lawyer Kert Stavorn said there was a risk of conflating an increase in self-represented claims with an increase in unmeritorious litigation.

“There is certainly a risk that the two can be conflated, and there is probably a degree of correlation,” he said.

“However, an increase in self-represented claims could itself tell us that there was a genuine need for access to justice in the first place. Whether it also leads to more unmeritorious litigation is a more nuanced issue.”

A person being advised that their proposed claim has little legal basis is one thing. A person with an arguable claim never obtaining that advice because representation is unaffordable is another.

Calderone also cautioned against overlooking what the technology can offer people who might otherwise struggle to pursue their rights.

“I am not suggesting that AI is wholly detrimental. It is no doubt assisting employees to better understand and pursue legitimate claims that they may otherwise have lacked the resources, knowledge or confidence to advance,” she said.

“That access to justice benefit is significant and should not be overlooked.”

For the vulnerable and marginalised communities the National Justice Project works with, Stavorn said existing barriers are already compounded by other forms of disadvantage.

“We think that the rise of generative AI in this space is in itself a testament to the systemic problems within the legal system,” he said.

“If anything, we feel it speaks to the desperation that vulnerable communities experience when dealing with the legal system.”

The potential of AI-assisted self-representation has been highlighted by the recent Fair Work dispute brought by computing academic Gregory David Baker, whose landmark AI-assisted Fair Work victory was previously reported by Lawyers Weekly.

Baker represented himself in a dispute over his casual employment status and ultimately succeeded in Application by Gregory David Baker to deal with a dispute about changing from casual employment [2026] FWC 3054.

His approach, however, was far removed from asking a free chatbot for legal advice. Baker subsequently described using multiple AI agents to help prepare and test his case, drawing on his own considerable technical expertise.

For Stutley, that qualification matters when successful cases are held up as evidence of AI’s access-to-justice potential.

“AI success stories have involved very sophisticated users of AI which have been proficient in training agents to achieve successful outcomes,” he said.

“This is far from the everyday user of a commercially available free version of generative AI.”

Information without judgement

Users need a degree of sophistication in AI to determine whether the legal information it produces is reliable.

“I think there is not enough awareness of how generative AI works in the general discourse,” Stavorn said.

“Generative AI is very much a people-pleaser in that it will try to give you the answer it thinks you want.”

The usefulness of the response can depend on the quality of the question, he said – presenting an obvious difficulty for people turning to AI precisely because they lack the knowledge they would otherwise obtain from a lawyer.

“Users looking to use it for self-representation are probably the kind of users who don’t have the depth of knowledge to ask those kinds of questions in the first place,” Stavorn said.

“That is the real dilemma we face in this space right now.”

Practitioners say that can affect more than the accuracy of a submission.

Stutley has seen AI contribute to unrealistic expectations about the strength or likely outcome of a claim, as well as reduced trust in practitioners and the process itself, as litigants instead place their trust in AI.

Calderone has similarly encountered applicants commencing or continuing proceedings because AI has suggested their claims are worth substantially more than they objectively are, or overstated their prospects of success.

“The challenge lies in ensuring that greater accessibility does not come at the expense of accuracy, procedural fairness, or the efficient resolution of disputes,” Calderone said.

“While AI has the potential to improve access to legal information, it must not be treated as a substitute for professional judgment, critical analysis and experience. Striking the right balance between access and reliability is essential.”

The burden on respondents

For respondents, even an ultimately unsuccessful claim must still be dealt with.

“Employer (respondents) are bearing a disproportionate burden,” O’Halloran said.

AI-generated material can be extensive and require careful analysis to identify incorrect authorities, misapplied legal tests or factual assertions that do not support the argument being made.

Stutley said that even legitimate issues can become buried within lengthy, repetitive submissions, increasing the time and cost required to identify what actually needs to be determined.

Recent decisions provide examples of how far those problems can go.

In Sadnan Khan v Aldi Pty Ltd [2026] FWC 3144, a self-represented applicant pursued an unfair dismissal claim despite falling three days short of the statutory minimum employment period. The case has previously been examined by Lawyers Weekly in an analysis of the risks of AI-driven self-representation.

Khan’s AI-generated submissions repeatedly addressed the wrong legal question. After the commission explained the problem in plain terms and warned of possible costs consequences, he continued filing AI-generated submissions before eventually conceding the point and discontinuing his application at the hearing.

Deputy president Michael Easton subsequently ordered him to pay $1,230 towards ALDI’s costs.

The consequences were considerably greater in Raghib v Stantec Australia (Costs) [2026] FCA 1415, where Justice Michael Wheelahan identified incoherent material and references to documents that did not exist, attributing some of the problems to the “indiscriminate use of artificial intelligence”. The applicant was ordered to pay $45,000 in costs.

Such orders remain the exception rather than the starting point in Fair Work litigation, where parties generally bear their own costs subject to specified statutory exceptions.

O’Halloran believes the existing mechanisms are ill-suited to the new environment.

“The current framework has some tools in theory, such as costs orders, jurisdictional objections, and the commission’s power to dismiss applications lacking merit, but none of these mechanisms appears to work in practice,” he said.

“Costs orders in Fair Work claims are rarely made, and that section of the Fair Work Act never contemplated AI slop being filed so readily, if at all!”

The pressure may also be affecting how disputes move through the commission.

Calderone said conciliations can feel shorter and more compressed than previously, at times leaving less opportunity for submissions on the merits, expectation management and meaningful dispute resolution.

“Where positions are not adequately tested at an early stage, more matters may continue through the system,” she said.

Calderone’s observation points in the same direction as other practitioners’ experiences, although there is not yet a commission data establishing that AI is systemically causing matters to settle less frequently.

Calderone is also seeing the effects before formal proceedings begin.

“We are also seeing an increase in workplace complaints, including whistleblower complaints. AI makes it considerably easier for individuals to prepare detailed allegations, chronologies and legal submissions,” she said.

“While accessibility is important, the ease of generating a complaint does not necessarily correlate with the strength of the underlying claim. Nevertheless, employers have legal and governance obligations to investigate and respond, often at substantial cost and with significant management distraction, particularly where the complaints reach senior executives and boards.”

Where complaints purport to be whistleblower complaints, or otherwise meet the requirements for whistleblower protection, Calderone said the risks increase.

“Whistleblowers must be afforded statutory protections, including in relation to confidentiality and detriment, irrespective of whether the substantive allegations are ultimately established,” she said.

“This significantly increases the risks and exposure for organisations and individuals involved in managing those complaints and the investigation process.”

Should the gate go back up?

With more claims able to enter the system without professional scrutiny, attention is turning to whether weak or misconceived matters should be identified earlier – and who should be responsible for doing so.

O’Halloran believes stronger early-stage merits assessment should be considered, together with changes to the costs jurisdiction for vexatious or misconceived AI-assisted claims.

He has also floated requiring litigants to file a certificate confirming they consulted a specialist lawyer when originally formulating their claim and did not rely entirely on AI.

The proposal would restore some professional scrutiny before a claim progresses through the system. It would also require a person who may have turned to AI because they could not access professional advice to obtain precisely that advice.

Stavorn said measures designed to deter weak claims risk disproportionately affecting people already facing significant barriers to justice.

“In our area of work, the clients we help are often from vulnerable and marginalised communities. They already face intersectional disadvantage,” he said.

“Where trauma or mental health conditions are involved, there is a strong likelihood that any additional barrier would have a compounding effect.”

The FWC has taken a different approach to regulating the technology itself.

From 20 October, parties who use generative AI to prepare documents in commission matters will be subject to new FWC requirements governing the use of AI in commission cases, including disclosure of when and how the technology was used and checks that documents are correct and relevant.

Rather than excluding AI from the process, the requirements seek to make the person using it accountable for what ultimately comes before the commission.

Building better pathways

For Stavorn, avoiding new barriers does not mean leaving self-represented litigants to navigate general-purpose AI tools without assistance.

Courts and tribunals could develop domain-specific AI grounded in authoritative information, he said, giving users a way to assess whether their case has merit and, if so, navigate the relevant processes and requirements.

The National Justice Project is pursuing a version of that approach through Hear Me Out, a free AI-powered complaints platform led by Stavorn.

The organisation said the tool uses verified data and guardrails to help users identify the appropriate complaint pathway, while a separate complaint-writing tool produces material specific to the relevant body.

“By doing this, we are trying to ensure that the complaints brought have merit and are submitted in a way that doesn’t add to the finite resources of that body,” Stavorn said.

Stutley said improving the quality and accessibility of authoritative legal information could serve a similar purpose.

The Fair Work Ombudsman, FWC, Safe Work Australia, and Australian Human Rights Commission already publish extensive guidance, he said, but there may be a case for investing further in those resources as AI becomes more widely used.

“Where authoritative guidance is clear, comprehensive and easy to access, there is a greater prospect that individuals using AI tools will receive information that is accurate, balanced and grounded in the relevant legal framework,” Stutley said.

The FWC’s own approach already acknowledges that litigants will continue to use the technology. Its new guidance does not simply warn users about generative AI; it provides resources designed to help them understand its benefits, risks and their responsibilities when using it.

“Ultimately, the objective should be to preserve the access-to-justice benefits that AI can provide, while ensuring that parties remain responsible for verifying the accuracy of the material they place before courts and tribunals,” Stutley said.

Beyond legal argument

AI-assisted legal submissions are one problem. AI-assisted evidence raises a more fundamental question about the integrity of the material on which courts and tribunals make decisions.

The FWC’s new requirements impose additional checks for witness statements and declarations, including that the evidence is based on the witness’s own knowledge, reflects their own words and is true to the best of their knowledge.

Stutley sees a significant risk if self-represented litigants instead begin using generative AI to create witness evidence.

“This does not protect access to justice, it erodes the integrity of the evidentiary record and, ultimately, the ability of courts and tribunals to determine the truth,” he said.

“Perhaps we are on the cusp of a return to viva voce evidence and less document-intensive trials.”

For now, the justice system is confronting a technology that is removing barriers that never served the same purpose.

Cost, complexity, and lack of legal knowledge have kept legitimate claimants out. Professional advice and merits assessment have kept some weak claims from proceeding. Generative AI is making it easier to bypass both.

The question is not whether those barriers should simply be rebuilt. It is what can replace the useful filtering they provided, without again putting justice beyond the reach of people who could never afford to get through them.