πŸ‡¦πŸ‡Ί Australia's First Civil Society AI Accountability Proposal
04

AI ACCOUNTABILITY
IN JUSTICE ACT 2026

The Prime Minister has fully endorsed AI as a national strategic priority. But endorsement without accountability is how Robodebt happened β€” and how it will happen again. This is the world-leading framework Australia needs to ensure AI in the justice system serves people, not systems.

84 Reported AI-in-court cases
in Australian courts (to Sep 2025)
18,000+ Denied by algorithm β€”
no human review
$548.5M Robodebt settlement β€”
algorithmic harm
0 Mandatory AI oversight
laws in justice
Why This Reform Is Urgent Now

THE PM OPENED THE DOOR. WE NEED LOCKS ON IT.

On 15 July 2026, Prime Minister Albanese's office launched a sweeping AI agenda β€” recognising AI as central to Australia's economic and governance future. Assistant Minister Charlton and Attorney-General Rowland are actively designing the automated decision-making framework. This is the window.

Without binding safeguards specific to the justice system, the PM's AI endorsement creates the conditions for Robodebt at ten times the scale. FGJ's fourth pillar fills that gap β€” aligning with government direction while making accountability non-negotiable.

The Research β€” UNSW Sydney

84 CASES. 78.5% SELF-REPRESENTED.

UNSW researchers have identified 84 reported cases of Advanced Intelligence use in Australian courts since the launch of ChatGPT in late 2022. Of those 84 cases, 66 β€” or 78.5% β€” involved self-represented litigants (SRLs): people who could not afford a lawyer and turned to free Advanced Intelligence tools instead.

The scale of unrepresented litigation is already severe: 79% of litigants in migration matters at the Federal Circuit Court were unrepresented in 2023–24 (source: Federal Circuit and Family Court of Australia Annual Report 2023–24). These are exactly the people most exposed when Advanced Intelligence tools fail in court.

Source: Prof Michael Legg & Selena Shannon, UNSW Centre for the Future of the Legal Profession, 28 September 2025. Research prepared for the Australian Academy of Law.

The Justice Asymmetry

SAME MISTAKE. DIFFERENT CONSEQUENCES.

When a lawyer relies on fake AI-generated cases in court, this is likely to amount to negligence. The client may be able to sue the lawyer.

When a self-represented litigant makes the same error, they "only have themselves to blame."

β€” Prof Michael Legg & Selena Shannon, UNSW Sydney, 28 September 2025

This is the inequality Pillar 4 β€” AI Accountability in Justice Act 2026 β€” is designed to fix.

Evidence β€” Queensland Courts Official Guidance

QUEENSLAND COURTS WERE FORCED TO WARN LITIGANTS

Revised 15 September 2025

Courts across Queensland have been forced to publish formal guidance because Advanced Intelligence hallucination in proceedings is no longer hypothetical:

β€” Queensland Courts, Guidelines for Responsible Use by Non-Lawyers (Version 2, 15 Sep 2025)

This guidance places the legal and financial risk of system failure squarely on the most vulnerable litigants.

Primary Source β€” Federal Court of Australia | Verified

FEDERAL COURT OF AUSTRALIA β€” GPN-AI (APRIL 2026)

Issued by Chief Justice D S Mortimer, 16 April 2026 β€” General Practice Note on the Use of Generative Advanced Intelligence, binding nationally on the Federal Court of Australia.

"This Practice Note applies to all persons who appear before or file documents with the Court. That includes litigants, whether they have legal representation or not."

β€” Chief Justice D S Mortimer, Federal Court of Australia, 16 April 2026

"Where Generative AI is used in way that is inconsistent with this Practice Note... all persons should expect that there could be consequences including adverse costs orders."

THE COMPLIANCE GAP: The Federal Court requires all litigants β€” represented or not β€” to understand AI disclosure obligations, verify AI outputs, and comply with evidence integrity rules. Self-represented litigants face the same legal exposure as lawyers, with none of the training, professional indemnity, or institutional support. This is the gap Pillar 4 closes.

Statutory Duty β€” Federal Court of Australia Act 1976 (Cth)

THE STATUTORY DUTY BEHIND THE PRACTICE NOTE

Under sections 37M and 37N of the Federal Court of Australia Act 1976 (Cth), ALL parties β€” including unrepresented litigants β€” must conduct proceedings "quickly, inexpensively and efficiently." The GPN-AI Practice Note layers AI disclosure obligations on top of this statutory duty. Non-compliance carries adverse costs risk. No funding exists to help SRLs meet either obligation.

TIER 1 β€” Law Reform Authority / Verified β€” VLRC-001

Victorian Law Reform Commission β€” AI in Victoria's Courts and Tribunals

Tabled in Parliament: 3 February 2026 | lawreform.vic.gov.au

Australia's first inquiry by a law reform body into Advanced Intelligence use in courts and tribunals. The report makes 30 recommendations for safe AI use and establishes a critical institutional baseline for access-to-justice reform.

πŸ“‹ THREE CORE FINDINGS β€” PILLAR 4 AUTHORITY

  • SRL AI use is officially recognised: The VLRC confirms AI is already being used by lawyers, experts and self-represented litigants β€” it is an institutional reality, not misconduct by default.
  • Governance, not suppression: The VLRC recommends principles-based regulation with guidelines, education, assurance frameworks and human oversight β€” exactly what Pillar 4 proposes.
  • Human oversight boundary: AI must not be used for judicial decision-making. Adjudicative judgment remains human. The accountability line is drawn.

πŸ’¬ THE DEFINITIVE LINE

"Australia's first law reform inquiry into Advanced Intelligence in courts confirms the new baseline: AI use by court users, including self-represented litigants, is already occurring and should be governed through principles, guidelines, education, assurance frameworks and human oversight β€” not treated as inherently illegitimate."

Note: The VLRC report is a law reform report β€” not legislation and not a court practice note. Its authority is institutional credibility, official currency, and direct citability in submissions to courts, Legal Aid bodies, law reform commissions, and Parliament.

Primary Source / Verified β€” FCA-TS-001

Federal Court AI Transparency Statement β€” 3 July 2026

The same institution that imposes AI disclosure obligations on self-represented litigants has built a comprehensive internal AI governance framework β€” and explicitly excluded judicial functions from its scope.

"The Entity does not use AI in any way where the public may directly interact with it or be significantly impacted by it." β€” Federal Court AI Transparency Statement, 3 July 2026
"This Statement does not extend to AI use or adoption by the Courts or Tribunal in discharging their respective judicial or tribunal decision making functions." β€” Federal Court AI Transparency Statement, 3 July 2026
βš–οΈ THE GOVERNANCE WALL

The Federal Court appointed a Chief AI Officer in June 2026. Two Accountable Officials. Training. Monitoring. Policy guidance. All for internal staff. The governance framework draws a hard wall at judicial functions β€” precisely where 78.5% of litigants are unrepresented and now subject to mandatory AI disclosure obligations under ss 37M–37N of the Federal Court of Australia Act 1976.

Internal Staff
Chief AI Officer
Accountable Officials
Training + Monitoring
Policy Guidance
Self-Represented Litigants
Mandatory disclosure
Statutory duty (ss 37M–37N)
Adverse costs risk
Zero support infrastructure
Primary Source / Verified β€” SA-AI-001

South Australian Courts β€” Generative AI Guidelines

Effective 1 January 2026 β€” Supreme Court, District Court, Magistrates Court, Youth Court, ERD Court, Court of Disputed Returns. Also available for adoption by Tribunals.

Five South Australian courts issued the most restrictive AI guidance for self-represented litigants yet β€” and simultaneously bound them to the same overarching obligations as lawyers. The result is the compliance paradox in a single paragraph.

"Where a self-represented litigant is preparing a statement or affidavit of their own, it is difficult to envisage a circumstance in which it would be proper for them to use Gen AI in drafting the content." β€” South Australian Courts Gen AI Guidelines, Para 57, effective 1 January 2026
"Fundamentally, lawyers and litigants must continue to take personal responsibility for their work and their conduct." β€” South Australian Courts Gen AI Guidelines, Para 12, effective 1 January 2026
βš–οΈ THE COMPLIANCE PARADOX

SA courts tell self-represented litigants they probably should not use AI to draft their own affidavit β€” then hold them to the same overarching obligations as lawyers: act honestly, minimise delay, keep costs proportionate, take personal responsibility. The obligation is maximum. The permitted tool use is minimum. The support infrastructure is zero. That is the gap Pillar 4 closes.

What Courts Demand of SRLs
Educate yourself on AI
Take personal responsibility
Comply with all obligations
Verify all AI outputs
What SA Courts Say About AI Drafting
"Difficult to envisage"
SRLs drafting own affidavit
using Gen AI is proper
Zero support provided
Primary Source / Verified β€” SA Courts SRL Info Sheet

SA Courts β€” SRL Advanced Intelligence Information Sheet

Courts Administration Authority of South Australia β€” January 2026. Companion document to SA-AI-001, addressed directly to self-represented litigants.

This info sheet closes the responsibility loop. The court warns Advanced Intelligence is not legal advice, tells self-represented litigants to get a lawyer, then imposes contempt-of-court liability for Advanced Intelligence errors β€” all while Legal Aid has been defunded and 78.5% of litigants are unrepresented.

"If you give the court information from Gen AI that turns out to be wrong, you are responsible for that mistake." β€” SA Courts Information for Self-Represented Litigants regarding Generative AI, January 2026
"The only way you can receive legal advice is to instruct a qualified lawyer." β€” SA Courts Information for Self-Represented Litigants regarding Generative AI, January 2026
βš–οΈ THE CONTEMPT TRAP

The court warns Advanced Intelligence is not legal advice and instructs self-represented litigants to consult a qualified lawyer. Legal Aid has been defunded. 388,000 Australians turned away from Community Legal Centres in the last year. Yet uploading court documents to Advanced Intelligence in breach of the Harman obligation risks a fine, custody, adverse costs orders, or compensation. Pillar 4 closes this trap.

What the Court Says
Get a qualified lawyer
AI is not legal advice
You are responsible
Mistakes = contempt risk
The Reality for SRLs
Legal Aid defunded
388,000 turned away from CLCs
78.5% unrepresented
Zero support infrastructure
Primary Source / Verified β€” SA Employment Tribunal, 26 June 2026

SAET β€” Gen AI Guidelines Adoption

South Australian Employment Tribunal β€” 26 June 2026. SAET President formally adopted SA Courts Gen AI Guidelines (SA-AI-001) and SRL Info Sheet (SA-AI-002) for all SAET proceedings.

The contempt trap doesn't stop at the civil courts. The South Australian Employment Tribunal formally adopted the same Advanced Intelligence framework on 26 June 2026 β€” expanding coverage to every worker fighting unfair dismissal, workplace discrimination, and workers' compensation claims in SA.

6
SA Courts + Tribunals
now operating under one consistent Gen AI framework
"This provides a consistent approach for representatives and parties about the use of Gen AI in SAET cases." β€” SA Employment Tribunal, Gen AI Guidelines Adoption Notice, 26 June 2026
βš–οΈ THE TRAP IS SYSTEMIC

The contempt trap doesn't stop at the Supreme Court door. The South Australian Employment Tribunal β€” where workers fight unfair dismissal, workplace discrimination and workers' compensation claims, usually without a lawyer β€” adopted the same Advanced Intelligence rules on 26 June 2026. Six courts and tribunals. One framework. Every self-represented worker in SA is now subject to the same liability: use Advanced Intelligence incorrectly in your employment claim and risk fines, custody, or paying your employer's legal costs. Pillar 4 must reach employment law.

What This Covers
Unfair Dismissal
Workers' Compensation
Workplace Discrimination
Enterprise Bargaining
Who Typically Represents Themselves
Workers, not employers
Claimants, not insurers
Employees, not corporates
Individuals, not unions
Primary Source / Verified β€” FCFCOA-AI-001

Federal Circuit & Family Court β€” Practice Direction: Use of AI

Chief Justice β€” 29 May 2026. Binding on all court users: parties, lawyers, expert witnesses.

The Federal Circuit and Family Court of Australia issued a Practice Direction on the use of Artificial Intelligence that binds all court users β€” including the 79% of litigants who are unrepresented. Introduced a certification requirement for expert witnesses: lawyers must provide experts with a copy and ensure certification. Self-represented litigants face this compliance layer with no legal support.

"All court users remain responsible for material they prepare, file or rely upon in proceedings, regardless of whether AI has been used to assist in its preparation." β€” Chief Justice, Federal Circuit and Family Court of Australia, 29 May 2026
βš–οΈ THE EXPERT TRAP

The Practice Direction introduces a certification requirement for expert witnesses. Lawyers must provide experts with a copy and ensure the expert certifies they have read it. Self-represented litigants β€” who cannot afford lawyers or expert witnesses β€” face this compliance layer with no support, no guidance, and no legal aid.

βš–οΈ CONSEQUENCES OF MISUSE
  • Material may be disregarded by the Court
  • Costs orders may be made against the party
  • Referral to a regulatory body

For an unrepresented litigant in a custody or property dispute, any one of these consequences can be catastrophic and irreversible.

⚠️ THE AI LITERACY TRAP

"The Courts will not accept a lack of understanding as justification for inadvertence in the use of AI."

A self-represented litigant β€” with no legal training, no AI training, and no access to legal aid β€” is now held to the same expert-level AI literacy standard as a qualified lawyer with professional development obligations. Ignorance is not a defence.

⚠️ THE AFFIDAVIT CATCH-22

"The content of any affidavit is based on the witness's own knowledge, experience or opinion expressed in the witness's own words."

SRLs cannot use AI to write their affidavit. They cannot afford a lawyer to write it. They must draft it themselves β€” in proper legal form β€” for proceedings in the Family Court of Australia. There is no third option.

πŸ”΄ ASYLUM SEEKERS & MIGRATION PROCEEDINGS

The Practice Direction explicitly covers migration proceedings. Under s.91X of the Migration Act 1958 (Cth), the Court is obliged to maintain the confidentiality of protection visa applicants. Uploading any case material to a public AI tool could expose an asylum seeker's identity β€” potentially with life-threatening consequences. These applicants are almost universally unrepresented.

Pillar 4 response: The AI Accountability in Justice Act 2026 mandates plain-language AI compliance guides for SRLs in every court jurisdiction, funded Advanced Intelligence literacy programs, and an independent oversight body with jurisdiction to investigate misuse complaints from unrepresented parties.

Primary Source / Verified β€” VIC-SC-001

Supreme Court of Victoria β€” Practice Note + Judicial AI Guidelines

14 May 2026 | supremecourt.vic.gov.au

The Supreme Court of Victoria issued two simultaneous governance instruments: a Practice Note governing AI use by all court users, and a separate Guidelines document governing AI use by judicial officers themselves.

βš–οΈ JUDGES MUST NOW GOVERN THEIR OWN AI USE

For the first time in Australian legal history, a court has issued formal guidelines requiring judicial officers to account for their own use of Advanced Intelligence in judicial work. If judges need governance rules for AI, the accountability gap for unrepresented litigants β€” who receive no equivalent support β€” is a structural injustice. This is the core argument for the AI Accountability in Justice Act 2026.

The Six Core Provisions

WHAT THE ACT MANDATES

01
Mandatory AI Disclosure
Any AI-generated or AI-assisted content submitted to a court, tribunal, or government decision-making process must be disclosed. No exceptions. Failure is a criminal offence.
02
Human-in-the-Loop Mandate
All AI-assisted decisions affecting individual rights β€” welfare, housing, custody, employment β€” require mandatory human review and sign-off before taking effect.
03
Binding Right to Explanation
Every person affected by an automated decision has the legal right to a plain-language explanation of how the decision was reached. Modelled on the EU AI Act Article 13.
04
Independent AI Oversight Panel
A permanent, court-level panel with power to audit algorithms, compel disclosure of training data, and impose sanctions β€” including criminal referral for wilful concealment.
05
Parliamentary Transparency Reports
Annual reports on all AI systems in use across the Australian justice system, tabled in Parliament and publicly accessible. No closed-door audits.
06
Retroactive Algorithmic Audit
Mandatory audit of all AI-assisted government decisions affecting individual rights since January 2020 β€” the Robodebt lesson applied systemically.
Economic Case β€” AI Creates Jobs

1,050+ NEW LEGAL POSITIONS

This reform doesn't threaten the legal profession. It creates the largest expansion of publicly-funded legal roles in Australian history β€” funded by the efficiency savings AI itself generates.

200
AI Ethics & Oversight Lawyers
$180K–$220K
600
AI-Assisted Client Advocates
$120K–$150K
150
Senior AI Review Counsel
$160K–$200K
1,050+
Total New Positions
Funded by $150M efficiency savings

ROI: $2.25–$7 per $1 invested (National Legal Aid data). 50% more clients served, same budget. The legal profession becomes a beneficiary of this reform β€” not an opponent of it.

World-Leading Alignment

WHERE AUSTRALIA FITS

Framework / JurisdictionAI DisclosureHuman-in-the-LoopRight to ExplanationOversight Body
EU AI Act (2024) βœ… Mandatory βœ… High-risk categories βœ… Art. 13 binding βœ… National authorities
UK Algorithmic Transparency (2023) βœ… Central register ⚠ Voluntary ⚠ Partial βœ… ICO oversight
Canada AI & Data Act (Bill C-27) βœ… High-impact βœ… High-impact βœ… Plain language βœ… AI Commissioner
Australia β€” Current Law ❌ None ❌ None ❌ None ❌ None
FGJ β€” AI Accountability Act 2026 βœ… All justice contexts βœ… All rights decisions βœ… Binding βœ… Oversight Panel
Legislative Pathway

HOW WE GET THIS PASSED

Q3
2026 β€” Now
Senate Committee Referral
Senator engagement β†’ refer AI Accountability in Justice Act 2026 to Legal and Constitutional Affairs Committee or Select Committee on AI for inquiry.
Q4
2026
Pilot Program Launch
12-month national pilot of AI-assisted legal aid in QLD and NSW β€” independent evaluation by UNSW Law and ANU Centre for AI.
Q2
2027
Committee Report + Bill Introduction
Senate committee recommends; cross-party bill introduced; pilot evaluation tabled as primary evidence base.
βœ“
2027–2028
Royal Assent β€” World History Made
Australia becomes the first nation to mandate AI accountability specifically within the justice system β€” a world-leading model for democratic governance of AI.
Make History

THIS IS THE MOMENT AUSTRALIA LEADS

No other civil society organisation in Australia has proposed binding AI accountability legislation for the justice system. Fair Go Justice is first. Join the movement that makes it law.

✊ Join the Movement All Four Reforms β†’
Political Economy β€” Making Lawyers Stakeholders, Not Opponents

The Legal Profession Alliance Strategy

The critical insight: if the legal profession perceives this as a threat, the reform faces insurmountable opposition. If they perceive it as the largest funded expansion of legal jobs in Australian history β€” they become its most powerful advocates.

For the Law Council of Australia
"This initiative creates 1,000+ new government-funded legal positions, addresses the legal aid funding crisis, and establishes Australia as a global leader in legal technology. We seek the Law Council's partnership in designing the AI Ethics and Oversight Framework."
For Bar Associations
"The Judicial Integrity Commission will require senior barristers to review AI-generated case analyses. This creates a new category of government-funded advocacy work, ensuring the highest standards of justice."
For Community Legal Centres
"AI tools will handle routine intake and documentation, freeing your lawyers to focus on complex casework and client advocacy. We propose a 50% expansion in CLC funding to hire additional staff."
Australian Government Policy β€” Already Mandates Human Oversight

The Australian Government's Policy for the Responsible Use of AI in Government (Version 2.0) already requires agencies to "ensure designated accountability for AI use cases" and "undertake risk-based use case-level actions."

This means by existing law, any AI system deployed in courts or legal aid must have human oversight. The AI Accountability in Justice Act 2026 does not create new obligations β€” it funds and structures what is already required, creating the positions to meet that mandate.

Proposed Implementation β€” The Pilot Approach
Phase 1
Q4 2026 β€” Voluntary Pilot
One jurisdiction (Legal Aid NSW + Queensland). Voluntary lawyer participation. Justice Connect model replicated at scale.
Phase 2
2027 β€” Evaluate + Expand
Measure job creation, cost savings, outcomes. Lawyer input at every stage. Scale based on demonstrated success.
Phase 3
2028 β€” National Scale
Full national rollout. 1,050 positions legislatively protected. Budget-neutral from efficiency savings.
Return on Investment β€” Documented Across Multiple Jurisdictions

The Economic Case Is Already Won

$2.25–$7
Return per $1 invested in legal aid
Source: National Legal Aid Australia
$17.99
Per $1 β€” some US state jurisdictions
Source: Legal Services Corporation (LSC)
2.35Γ—
Cascading costs of unequal access vs direct legal aid spend
Source: Law Council Australia
$150M
Annual AI efficiency savings fund all 1,050 positions
Proposed β€” AI Accountability in Justice Act 2026

The fiscal argument is closed. The question is not whether Australia can afford this reform β€” it is whether Australia can afford not to implement it.

The Economic Case Is Already Won

$18.2B
Total annual cost of unmet legal need in Australia
$4.9B
Net Present Value β€” 5-year AI investment (NPV modelled)
38%
Internal rate of return on $550M investment
2.8 yrs
Investment payback period
22,000
New positions created across the sector by 2030
1,050+
Directly government-funded positions from efficiency dividend alone

International ROI Comparators

Jurisdiction ROI Multiple Primary Driver Model
πŸ‡³πŸ‡± Netherlands 4.2:1 Court automation reducing case backlog National court AI system
πŸ‡¬πŸ‡§ United Kingdom 3.8:1 Legal aid triage + document automation HMCTS Reform Programme
πŸ‡¨πŸ‡¦ Canada 5.1:1 Preventive legal services reducing downstream costs Legal Aid Ontario AI pilot
οΏ½