The UNSW Centre for the Future of the Legal Profession (CFLP) has released a landmark analysis of how Generative AI is appearing in court and tribunal matters, and where “fake law” is creeping into filings. The report tracks a rapidly growing body of cases across multiple jurisdictions and offers a practical view of what courts are seeing, how they are responding, and the implications for legal teams, clients and self-represented litigants.
You can explore the CFLP research hub here: UNSW Law & Justice – Centre for the Future of the Legal Profession, and the interactive publication here: Generative AI, Fake Law and Fallout (flipping book).
Key Summary
- Scope: 520 GenAI-related matters identified across 10 jurisdictions (Jan 2023–Sep 2025).
- Australia: 87 cases — the second-highest volume after the US — with the majority involving self-represented litigants (SRLs).
- Recurring issues: fake/incorrect citations, flawed reasoning, procedurally defective documents and prolix “flooding” of filings.
- Court response: practical guidance, disclosure expectations where appropriate, and a spectrum of case-management measures that differentiate inadvertent misuse from serious or repeated breaches.
What the CFLP Dataset Covers
The report aggregates publicly available court and tribunal material across ten jurisdictions from January 2023 to September 2025. It focuses on instances where Generative AI appears explicitly, such as through party submissions, judicial remarks, practice notes, or related procedural events. As the authors note, this is almost certainly the “tip of the iceberg”: many matters never mention the underlying tools in written reasons, and informal use may not be visible at all.
At a glance: a total of 520 matters were identified for the period. Australia accounts for 87, clustering in jurisdictions and lists where SRLs are common. The majority of identified decisions fall in 2025, reflecting both a genuine uptick in usage and better visibility as courts begin to call out poor citation practices and issue guidance.
The Four Recurring ‘Fake Law’ Risk Themes
1) Fake or Incorrect Authorities
Courts continue to encounter hallucinated cases, misquoted provisions and mismatched citations. The pattern is familiar: a user prompts a general-purpose model, copies results directly into submissions, and fails to verify using an official source (authorised reports, legislation databases, or court websites). When detected, courts are increasingly explicit about the need to excise or disregard such material to prevent further propagation.
2) Flawed Legal Reasoning
Generative text can sound persuasive while quietly misapplying tests, reversing onus, or collapsing multi-limb analyses. The report highlights that unverified reasoning is a greater systemic risk than single fake citations: it can distort entire submissions, wasting court time and obscuring the real issues in dispute.
3) Incorrect or Defective Documents
From wrong form types to procedural missteps and missing annexures, the dataset shows repeated instances where AI-generated documents do not meet basic filing requirements. This compounds workload for registries and judges, and can prejudice parties who rely on the documents being accepted as lodged.
4) Prolix Filings (Document ‘Flooding’)
The ease of producing long, repetitive text has led to over-length, unfocused submissions. Courts respond by directing parties back to the live issues, enforcing page limits, and—where appropriate—using costs or procedural powers to manage misuse that burdens the other side.
Australia: Who’s Using AI — and How Courts Are Responding
Of the Australian matters catalogued, a substantial majority involve self-represented litigants. Courts and tribunals most exposed to SRL filings see the volume first: small-claims jurisdictions, commissions and lower courts. However, the report also notes cases involving legal practitioners and occasional judicial remarks about limited, careful use of AI for administrative tasks.
Australian courts are converging around a practical stance:
- Verification and accuracy are non-negotiable; fake or unverifiable authorities are disregarded and, at times, redacted in published reasons.
- Disclosure expectations are evolving via practice notes and guidance; where required, parties are reminded that disclosing tool use does not excuse a lack of verification.
- Proportionate case management distinguishes inadvertent SRL mistakes from serious or repeated practitioner breaches, with measures ranging from educative directions to costs orders and regulatory referrals.
For a broader look at how the judiciary’s approach has shifted through 2025, see our earlier piece: AI in the Courts: What’s Changed Since July? Judges Next?
Conclusion
The CFLP’s 2025 report paints a balanced picture. Generative AI is not an existential threat to the justice system, but it can amplify old problems in new ways. The common denominator in every risk theme is verification. Courts are building sensible guardrails; legal teams that pair AI with disciplined source-checking, concise drafting and clear supervision are best placed to benefit from the technology without importing “fake law” into the record.
Disclaimer: This blog is a general overview and should not be construed as professional legal, financial or medical advice.
FAQs
- What does the report mean by “fake law”?
The term covers problems like hallucinated cases, misquoted provisions and fabricated citations that slip into filings when AI outputs aren’t verified against authoritative sources.
- Are Australian courts banning Generative AI?
No. Courts emphasise duties of candour and accuracy. Some practice notes encourage disclosure and mandate verification; misuse is dealt with through proportionate case management.
- Do practitioners need to disclose AI use?
Where practice notes or directions require it, yes. Disclosure does not replace verification — sources must still be checked using official reports and legislation databases.
- What’s the most common failure courts see?
Unverified authorities: often persuasive-sounding text with citations that don’t exist or don’t match the proposition. Courts are increasingly explicit about rejecting or redacting such material.
