--- title: "Class Actions / Representative Proceedings" url: "https://ailawframework.com/practice/class-actions-representative-proceedings" type: "practice_area" site: "AI Law Framework" id: "A7" task_count: 15 last_reviewed: "2026-10-01" --- # Class Actions / Representative Proceedings > Representative proceedings: viability, class definition, discovery, notices and settlement. AI organises documents and drafts; the lawyer protects group members, whose rights a single error can affect. Canonical page: https://ailawframework.com/practice/class-actions-representative-proceedings Last reviewed: 2026-10-01 Site index for agents: https://ailawframework.com/llms.txt Also covers: Representative proceedings; group proceedings; Part IVA proceedings (Federal Court); Division 9.3 proceedings (Supreme Court of Victoria); Pt 10 proceedings (Supreme Court NSW); group member litigation; securities class actions; consumer class actions; mass tort proceedings ## Key facts - Tasks: 15 - Green: 0, Yellow: 7, Red: 8 - Levels: Green is AI drafts and the lawyer checks; Yellow is lawyer and AI draft together; Red is the lawyer writes and AI may only check, clarify or collect data. ## Principles - **Define the matter**: Confirm the client’s instructions, jurisdiction, intended output and the information available before dividing the work. - **Verify the authorities**: Open each relied-on authority independently. Confirm its existence, current status, proposition and any quotation or pinpoint reference. - **Preserve the witness’s evidence**: Confirm each factual assertion with the witness and original record. Preserve the witness’s own knowledge and words, and check the court’s rules for this document. ## Checklist - Open each relied-on authority independently. Confirm its existence, current status, proposition and any quotation or pinpoint reference. - Confirm each factual assertion with the witness and original record. Preserve the witness’s own knowledge and words, and check the court’s rules for this document. - Trace every relied-on figure to its source, confirm dates and assumptions, and independently recalculate totals or derived amounts. - Calculate operative dates independently from the current rule and actual triggering event. Check exceptions and record the verified deadline. - Check the current court or regulator requirements, prescribed form and attachments for the intended document and filing. - Review operative clauses in the complete document, including definitions, cross-references, schedules and inconsistencies with the instructions. - Confirm that the wording reflects the client’s instructions and contains no unintended admission, concession or commitment. Check the recipient and tone. - Confirm that the actual tool, feature and processing terms permit use of this information, including access, retention, training and disclosure conditions. - Check the specialist’s assumptions, qualifications and limits. Obtain the relevant expert or specialist review before relying on a technical conclusion. - Use an appropriately qualified person to verify translation and meaning, and confirm any certification required for the intended use. - Verify dates, people and sequence against contemporaneous records. Preserve uncertainty and identify gaps or conflicting accounts. - Preserve original files and relevant metadata. Document provenance and handling, and seek forensic review where authenticity needs to be resolved. - Verify the applicable jurisdiction and the version of each relevant law, rule or instrument in force at the material time. - Resolve exceptions and record the responsible lawyer’s approval of the exact output and intended use. ## Restrictions - Confirm firm, client and applicable court or professional restrictions before using any AI feature. ## Jurisdiction note Apply this guidance to the actual matter and jurisdiction. Confirm current requirements before relying on an example. ## Basis Practice guidance supports planning; the responsible lawyer confirms the scope, evidence and review requirements for the matter. Also covers: Representative proceedings; group proceedings; Part IVA proceedings (Federal Court); Division 9.3 proceedings (Supreme Court of Victoria); Pt 10 proceedings (Supreme Court NSW); group member litigation; securities class actions; consumer class actions; mass tort proceedings Class actions aggregate individual claims into a single representative proceeding, governed by Part IVA of the Federal Court of Australia Act 1976 (Cth) at federal level and cognate state regimes (Pt 10 Supreme Court Act 1986 (Vic); Pt 10 CPA (NSW); Part 9A District Court Act 1973 (NSW)). The AI risk profile is exceptionally high: proceedings are factually dense, procedurally intricate, involve large volumes of confidential group-member data, require court-approved notices and settlements, and generate binding outcomes for thousands of people who never instruct a lawyer directly. Three structural amplifiers elevate risk above ordinary litigation: (1) every hallucinated authority, misdescribed group definition, or incorrect limitation date can extinguish the rights of the entire class with no individual recovery mechanism; (2) the volume and velocity of AI use in class action workflows (document review, group-member communications, loss modelling) creates multiple simultaneous policy-theatre vectors where informal AI use bypasses governance controls; and (3) GPN-AI (Federal Court, 16 April 2026) imposes mandatory disclosure and verification obligations that expose the responsible lawyer to personal adverse costs orders if AI outputs are filed without the required verification steps. ## Limits on AI assistance Red lines — never in this area ## Limits on AI assistance AI must never generate or finalise opt-out notices, registration notices, soft closure notices, or any group-member notice that will be distributed without complete lawyer re-drafting and court approval — court approval under s 33X FCA Act (and state equivalents) is mandatory; distributing an AI-generated unapproved notice is a procedural irregularity that can invalidate opt-outs filed in response. AI must never independently calculate or certify a settlement distribution entitlement for any group member — amounts paid to group members are court-approved and any error in an approved scheme has no mechanism for informal correction. AI must never be the sole source of authority verification in any document filed with a court — verified hallucination rates of 17-33% on specialist legal AI tools (Magesh et al.) and up to 91% on general LLMs (Chelli et al.) mean that every citation must be checked by a lawyer against the actual judgment text before filing; Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (indemnity costs personally against solicitor), Handa & Mallick [2024] FedCFamC2F 957 (referral to VLSBC; subsequent two-year supervised-practice condition), and Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 95 (referral to NSW OLSC) are live Australian disciplinary precedents applying equally to class action filings. AI must never process identifiable group-member PII (names, trading data, policy data, health records, registration data) through a public or uncontracted LLM — this breaches APPs 3, 6, and 11 under the Privacy Act 1988 (Cth), likely constitutes a notifiable data breach reportable to the OAIC under Part IIIC, and may constitute a 'serious invasion of privacy' under the statutory tort introduced by the Privacy and Other Legislation Amendment Act 2024 (Cth) (commenced 10 June 2025), exposing the firm to individual tortious claims from group members. AI must never draft, generate, or substantively contribute to the content of a lead plaintiff's affidavit, group-member witness statement, or any affidavit in class action proceedings — GPN-AI (Federal Court, April 2026) requires affidavits to reflect the witness's own recollection and knowledge; NSW SC Gen 23 imposes an absolute prohibition on AI-generated affidavit content without prior court leave. AI must never generate content within an expert report without prior court leave — GPN-AI requires that expert reports reflect the expert's own reasoning; GPN-EXPT (Harmonised Expert Witness Code of Conduct) imposes an overriding duty to the court that cannot be discharged through AI-generated content; NSW SC Gen 23 prohibits AI-generated expert report content without leave. AI must never draft a s 33V settlement approval submission or settlement distribution scheme without complete human re-drafting and independent verification — these documents bind thousands of non-participating group members and courts apply intensive scrutiny to both litigation risk assessments and fee structures. AI must never embed a solicitor percentage fee (common fund order in favour of the law firm) in any distribution model for Federal Court proceedings — Kain v R&B Investments Pty Ltd [2025] HCA 28 (7-0) held this impermissible as contravening LPUL NSW s 183; a model producing such output will produce a settlement approval submission the court will reject. AI must never determine the scope of privilege claims over discovered documents — inadvertent waiver of a tranche of privileged documents through AI misclassification cannot be remedied by clawback in many class action circumstances and the volume makes practical recovery impossible. AI must never advise on or compute limitation period expiry dates for class actions without independent lawyer verification — limitation periods interact with s 33ZE FCA Act (suspension for group members), applicable state Limitation Acts, and any tolling arguments in non-obvious ways; an AI error could extinguish the rights of the entire class. AI must never generate, fabricate, or reconstruct source documents (board minutes, ASX announcements, product disclosures, trading records) for use in pleadings, court books, or expert instructions — even inadvertent inclusion of an AI-generated document as if it were a discovered document is a contempt risk and a breach of overarching obligations under s 37N FCA Act. - AI must never generate or finalise opt-out notices, registration notices, soft closure notices, or any group-member notice that will be distributed without complete lawyer re-drafting and court approval — court approval under s 33X FCA Act (and state equivalents) is mandatory; distributing an AI-generated unapproved notice is a procedural irregularity that can invalidate opt-outs filed in response. - AI must never independently calculate or certify a settlement distribution entitlement for any group member — amounts paid to group members are court-approved and any error in an approved scheme has no mechanism for informal correction. - AI must never be the sole source of authority verification in any document filed with a court — verified hallucination rates of 17-33% on specialist legal AI tools (Magesh et al.) and up to 91% on general LLMs (Chelli et al.) mean that every citation must be checked by a lawyer against the actual judgment text before filing; Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (indemnity costs personally against solicitor), Handa & Mallick [2024] FedCFamC2F 957 (referral to VLSBC; subsequent two-year supervised-practice condition), and Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 95 (referral to NSW OLSC) are live Australian disciplinary precedents applying equally to class action filings. - AI must never process identifiable group-member PII (names, trading data, policy data, health records, registration data) through a public or uncontracted LLM — this breaches APPs 3, 6, and 11 under the Privacy Act 1988 (Cth), likely constitutes a notifiable data breach reportable to the OAIC under Part IIIC, and may constitute a 'serious invasion of privacy' under the statutory tort introduced by the Privacy and Other Legislation Amendment Act 2024 (Cth) (commenced 10 June 2025), exposing the firm to individual tortious claims from group members. - AI must never draft, generate, or substantively contribute to the content of a lead plaintiff's affidavit, group-member witness statement, or any affidavit in class action proceedings — GPN-AI (Federal Court, April 2026) requires affidavits to reflect the witness's own recollection and knowledge; NSW SC Gen 23 imposes an absolute prohibition on AI-generated affidavit content without prior court leave. - AI must never generate content within an expert report without prior court leave — GPN-AI requires that expert reports reflect the expert's own reasoning; GPN-EXPT (Harmonised Expert Witness Code of Conduct) imposes an overriding duty to the court that cannot be discharged through AI-generated content; NSW SC Gen 23 prohibits AI-generated expert report content without leave. - AI must never draft a s 33V settlement approval submission or settlement distribution scheme without complete human re-drafting and independent verification — these documents bind thousands of non-participating group members and courts apply intensive scrutiny to both litigation risk assessments and fee structures. - AI must never embed a solicitor percentage fee (common fund order in favour of the law firm) in any distribution model for Federal Court proceedings — Kain v R&B Investments Pty Ltd [2025] HCA 28 (7-0) held this impermissible as contravening LPUL NSW s 183; a model producing such output will produce a settlement approval submission the court will reject. - AI must never determine the scope of privilege claims over discovered documents — inadvertent waiver of a tranche of privileged documents through AI misclassification cannot be remedied by clawback in many class action circumstances and the volume makes practical recovery impossible. - AI must never advise on or compute limitation period expiry dates for class actions without independent lawyer verification — limitation periods interact with s 33ZE FCA Act (suspension for group members), applicable state Limitation Acts, and any tolling arguments in non-obvious ways; an AI error could extinguish the rights of the entire class. - AI must never generate, fabricate, or reconstruct source documents (board minutes, ASX announcements, product disclosures, trading records) for use in pleadings, court books, or expert instructions — even inadvertent inclusion of an AI-generated document as if it were a discovered document is a contempt risk and a breach of overarching obligations under s 37N FCA Act. ## Jurisdiction-specific considerations Australian traps ## Jurisdiction-specific considerations GPN-AI (Federal Court, effective 16 April 2026): the primary binding instrument for AI use in Federal Court class actions. Mandates disclosure in any court document (pleadings, submissions, affidavits, chronologies, lists of documents) of where and how AI was used; the disclosure must appear at the start of the document and must identify the specific tool used, where in the document it was applied, and how — a blanket footer does not satisfy this requirement. Requires the responsible lawyer to verify facts are provable, authorities exist and are accurate, evidence is admissible, and discovery obligations are met. Non-compliance: adverse costs orders and professional repercussions are stated consequences. GPN-CA (Class Actions Practice Note, Federal Court): governs case management of class actions including competing proceedings, opt-out and notice processes, settlement approval procedures, and the Federal Court-Victoria Supreme Court joint protocol for multi-jurisdictional proceedings. The Federal Court's sample opt-out notice template (published on the Federal Court website pursuant to GPN-CA) is the required baseline — AI-generated notices must be checked against that template. AI-assisted notice or settlement documents must comply with GPN-CA content requirements independently of whether GPN-AI disclosure has been made. GPN-EXPT (Expert Evidence Practice Note, Federal Court) and Harmonised Expert Witness Code of Conduct (Annexure A): experts in class actions (forensic economists, share market experts, actuaries) have an overriding duty to the court and must ensure their reports reflect their own reasoning. AI-drafted expert instructions must not embed conclusions that compromise independence. GPN-AI additionally requires disclosure where AI informed material upon which an expert relied. NSW Supreme Court Practice Note SC Gen 23 (Generative AI, effective 21 November 2024, revised 28 January 2025): imposes an absolute prohibition on using generative AI to generate the content of affidavits, witness statements, character references, or expert reports without prior court leave in NSW Supreme Court proceedings. Applies to NSW state class actions (Pt 10 CPA (NSW)). Differs materially from GPN-AI in that it is prohibitory, not merely disclosure-based. Queensland and Western Australia have equivalent practice notes; lawyers must identify the applicable court regime before any filing. s 33X FCA Act (and state equivalents: s 178 CPA NSW): court approval of the form and content of all group-member notices is mandatory — not merely a best practice. AI drafts must go through the mandatory approval process. Distributing an unapproved AI-generated notice is a procedural irregularity that can invalidate opt-outs filed in response. Any post-approval editing of an approved notice form, even minor, requires a further court application. s 33V FCA Act: settlement must be approved by the court as fair and reasonable. Courts have increased scrutiny of distribution scheme design, litigation-funder and solicitor fee structures. The ASIC v Noumi Limited (No 5) [2025] FCA 1524 precedent established that ASIC pecuniary penalties can be transferred to a class action settlement distribution scheme — AI distribution models must check for and accommodate this mechanism where parallel ASIC proceedings exist. Kain v R&B Investments Pty Ltd [2025] HCA 28 (6 August 2025, 7-0): the High Court held that solicitors' common fund orders (SCFOs) are impermissible in Federal Court proceedings because they contravene s 183 of the Legal Profession Uniform Law (NSW) (and equivalent state provisions). The ruling was anchored to LPUL NSW s 183; firms in other jurisdictions should obtain independent advice on whether equivalent Uniform Law provisions produce the same result. Any AI-generated distribution model embedding a percentage solicitor fee from the class fund is producing a legally incorrect output for Federal Court proceedings. Lendlease Corporation Limited v Pallas [2025] HCA 19: the High Court confirmed (in all four judgments) that both Federal and state courts have power to make soft class closure orders. The decision resolved an inconsistency between Federal and NSW Supreme Court positions. Soft closure orders are now a routine pre-mediation mechanism in both jurisdictions. AI-drafted pre-mediation notices must accurately describe that unregistered group members may lose the right to benefit from settlement while remaining bound by any judgment — misstating this relationship generates satellite litigation over whether affected group members retain settlement rights. Continuous disclosure reform — s 674A Corporations Act (Cth): the knowledge-based continuous disclosure standard introduced by the Treasury Laws Amendment (2021 Measures No. 1) Act 2021 (Cth) (in force May 2021, subsequently extended) applies to conduct post-May 2021. AI trained on pre-amendment law generates incorrect liability analyses for this period. The two-standard landscape (strict liability under s 674 for pre-May 2021 conduct; knowledge-based under s 674A for post-May 2021 conduct) is a recurring AI failure mode in securities class action pleadings. Zonia Holdings Pty Ltd v Commonwealth Bank of Australia Ltd [2024] FCA 477 (affirmed on appeal [2024] FCAFC): Australian courts will not strip out confounding information from share price movements to isolate the impact of the alleged non-disclosure — the onus is on shareholders to isolate the relevant price movement through expert evidence. AI event study tools that apply standard US-style confounding-adjustment methodology produce causation and loss analyses inconsistent with current Federal Court requirements. This affects both expert instructions and settlement quantum modelling. Privacy Act 1988 (Cth) as amended by the Privacy and Other Legislation Amendment Act 2024 (Cth) (royal assent 10 December 2024; statutory tort commenced 10 June 2025): APPs 3, 6, and 11 apply to group-member PII (collection limitation, secondary use restriction, security). The 2024 amendments introduced a statutory tort of serious invasion of privacy (misuse of personal information) that gives individual group members a cause of action in tort against the firm. Uploading group-member registration datasets, trading data, or health records to a public AI platform is a notifiable data breach reportable to the OAIC under Part IIIC and may independently constitute a serious invasion of privacy. The OAIC has strengthened enforcement powers under the 2024 amendments. Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (22 April 2025): Federal Court ordered indemnity costs personally against the solicitor after AI-generated hallucinated citations appeared in court documents. Handa & Mallick [2024] FedCFamC2F 957 (19 July 2024): court ordered referral to VLSBC after solicitor filed submissions with non-existent authorities generated by AI; the subsequent Victorian Legal Services Board disciplinary proceeding imposed a two-year supervised-practice condition. Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 95: solicitor referred to NSW OLSC for filing submissions with AI-generated non-existent citations and quotes. These are the controlling Australian authorities on consequences — they apply with equal force to class action filings. Corporations Regulations 2001 (Cth) reg 7.6.01AB and ASIC Regulatory Guide 248: litigation funders operating in Australian class actions are currently exempt from AFSL and managed investment scheme requirements (2022 legislative exemption, extended by ASIC to end of 2029) but must maintain adequate practices for managing conflicts of interest under reg 7.6.01AB. ASIC's current guidance is RG 248 (updated December 2024). AI-generated conflict mapping for funder compliance must be verified against the actual regulatory text and RG 248 — AI may conflate the former AFSL requirements (repealed by the 2022 exemption) with the continuing conflict management obligations that remain in force. s 37N and s 37M FCA Act (overarching obligations): the responsible lawyer personally certifies on every filed document that overarching obligations are met. These obligations apply to all parties and lawyers, are non-delegable, and cannot be discharged by reference to AI output. GPN-AI expressly incorporates these obligations into the verification requirements for AI-assisted court documents. - GPN-AI (Federal Court, effective 16 April 2026): the primary binding instrument for AI use in Federal Court class actions. Mandates disclosure in any court document (pleadings, submissions, affidavits, chronologies, lists of documents) of where and how AI was used; the disclosure must appear at the start of the document and must identify the specific tool used, where in the document it was applied, and how — a blanket footer does not satisfy this requirement. Requires the responsible lawyer to verify facts are provable, authorities exist and are accurate, evidence is admissible, and discovery obligations are met. Non-compliance: adverse costs orders and professional repercussions are stated consequences. - GPN-CA (Class Actions Practice Note, Federal Court): governs case management of class actions including competing proceedings, opt-out and notice processes, settlement approval procedures, and the Federal Court-Victoria Supreme Court joint protocol for multi-jurisdictional proceedings. The Federal Court's sample opt-out notice template (published on the Federal Court website pursuant to GPN-CA) is the required baseline — AI-generated notices must be checked against that template. AI-assisted notice or settlement documents must comply with GPN-CA content requirements independently of whether GPN-AI disclosure has been made. - GPN-EXPT (Expert Evidence Practice Note, Federal Court) and Harmonised Expert Witness Code of Conduct (Annexure A): experts in class actions (forensic economists, share market experts, actuaries) have an overriding duty to the court and must ensure their reports reflect their own reasoning. AI-drafted expert instructions must not embed conclusions that compromise independence. GPN-AI additionally requires disclosure where AI informed material upon which an expert relied. - NSW Supreme Court Practice Note SC Gen 23 (Generative AI, effective 21 November 2024, revised 28 January 2025): imposes an absolute prohibition on using generative AI to generate the content of affidavits, witness statements, character references, or expert reports without prior court leave in NSW Supreme Court proceedings. Applies to NSW state class actions (Pt 10 CPA (NSW)). Differs materially from GPN-AI in that it is prohibitory, not merely disclosure-based. Queensland and Western Australia have equivalent practice notes; lawyers must identify the applicable court regime before any filing. - s 33X FCA Act (and state equivalents: s 178 CPA NSW): court approval of the form and content of all group-member notices is mandatory — not merely a best practice. AI drafts must go through the mandatory approval process. Distributing an unapproved AI-generated notice is a procedural irregularity that can invalidate opt-outs filed in response. Any post-approval editing of an approved notice form, even minor, requires a further court application. - s 33V FCA Act: settlement must be approved by the court as fair and reasonable. Courts have increased scrutiny of distribution scheme design, litigation-funder and solicitor fee structures. The ASIC v Noumi Limited (No 5) [2025] FCA 1524 precedent established that ASIC pecuniary penalties can be transferred to a class action settlement distribution scheme — AI distribution models must check for and accommodate this mechanism where parallel ASIC proceedings exist. - Kain v R&B Investments Pty Ltd [2025] HCA 28 (6 August 2025, 7-0): the High Court held that solicitors' common fund orders (SCFOs) are impermissible in Federal Court proceedings because they contravene s 183 of the Legal Profession Uniform Law (NSW) (and equivalent state provisions). The ruling was anchored to LPUL NSW s 183; firms in other jurisdictions should obtain independent advice on whether equivalent Uniform Law provisions produce the same result. Any AI-generated distribution model embedding a percentage solicitor fee from the class fund is producing a legally incorrect output for Federal Court proceedings. - Lendlease Corporation Limited v Pallas [2025] HCA 19: the High Court confirmed (in all four judgments) that both Federal and state courts have power to make soft class closure orders. The decision resolved an inconsistency between Federal and NSW Supreme Court positions. Soft closure orders are now a routine pre-mediation mechanism in both jurisdictions. AI-drafted pre-mediation notices must accurately describe that unregistered group members may lose the right to benefit from settlement while remaining bound by any judgment — misstating this relationship generates satellite litigation over whether affected group members retain settlement rights. - Continuous disclosure reform — s 674A Corporations Act (Cth): the knowledge-based continuous disclosure standard introduced by the Treasury Laws Amendment (2021 Measures No. 1) Act 2021 (Cth) (in force May 2021, subsequently extended) applies to conduct post-May 2021. AI trained on pre-amendment law generates incorrect liability analyses for this period. The two-standard landscape (strict liability under s 674 for pre-May 2021 conduct; knowledge-based under s 674A for post-May 2021 conduct) is a recurring AI failure mode in securities class action pleadings. - Zonia Holdings Pty Ltd v Commonwealth Bank of Australia Ltd [2024] FCA 477 (affirmed on appeal [2024] FCAFC): Australian courts will not strip out confounding information from share price movements to isolate the impact of the alleged non-disclosure — the onus is on shareholders to isolate the relevant price movement through expert evidence. AI event study tools that apply standard US-style confounding-adjustment methodology produce causation and loss analyses inconsistent with current Federal Court requirements. This affects both expert instructions and settlement quantum modelling. - Privacy Act 1988 (Cth) as amended by the Privacy and Other Legislation Amendment Act 2024 (Cth) (royal assent 10 December 2024; statutory tort commenced 10 June 2025): APPs 3, 6, and 11 apply to group-member PII (collection limitation, secondary use restriction, security). The 2024 amendments introduced a statutory tort of serious invasion of privacy (misuse of personal information) that gives individual group members a cause of action in tort against the firm. Uploading group-member registration datasets, trading data, or health records to a public AI platform is a notifiable data breach reportable to the OAIC under Part IIIC and may independently constitute a serious invasion of privacy. The OAIC has strengthened enforcement powers under the 2024 amendments. - Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (22 April 2025): Federal Court ordered indemnity costs personally against the solicitor after AI-generated hallucinated citations appeared in court documents. Handa & Mallick [2024] FedCFamC2F 957 (19 July 2024): court ordered referral to VLSBC after solicitor filed submissions with non-existent authorities generated by AI; the subsequent Victorian Legal Services Board disciplinary proceeding imposed a two-year supervised-practice condition. Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 95: solicitor referred to NSW OLSC for filing submissions with AI-generated non-existent citations and quotes. These are the controlling Australian authorities on consequences — they apply with equal force to class action filings. - Corporations Regulations 2001 (Cth) reg 7.6.01AB and ASIC Regulatory Guide 248: litigation funders operating in Australian class actions are currently exempt from AFSL and managed investment scheme requirements (2022 legislative exemption, extended by ASIC to end of 2029) but must maintain adequate practices for managing conflicts of interest under reg 7.6.01AB. ASIC's current guidance is RG 248 (updated December 2024). AI-generated conflict mapping for funder compliance must be verified against the actual regulatory text and RG 248 — AI may conflate the former AFSL requirements (repealed by the 2022 exemption) with the continuing conflict management obligations that remain in force. - s 37N and s 37M FCA Act (overarching obligations): the responsible lawyer personally certifies on every filed document that overarching obligations are met. These obligations apply to all parties and lawyers, are non-delegable, and cannot be discharged by reference to AI output. GPN-AI expressly incorporates these obligations into the verification requirements for AI-assisted court documents. ## Risks to address How it goes wrong ## Risks to address Hallucinated or misquoted judicial authority in pleadings or submissions: the post-Myer securities class action landscape, the Zonia Holdings causation ruling, and the 2021 continuous disclosure reforms mean the law changed materially in 2021-2024; AI trained on older data cites cases for superseded propositions or generates plausible-sounding non-existent decisions — directly witnessed in Australian proceedings with personal costs consequences (Murray [2025], Handa [2024], Valu [2025]). Incorrect continuous disclosure version applied: AI applying the pre-2021 strict liability standard (s 674) to conduct occurring after the knowledge-based amendment to s 674A was enacted, or applying the temporary COVID-era regime to standard matters, producing a legally incorrect liability framework in the statement of claim that collapses at trial or summary judgment. Group definition that fails the s 33C ascertainability test: AI-generated definitions using language importing individual merits assessment (e.g., 'group members who suffered loss') do not satisfy the requirement that membership be objectively ascertainable — a struck-out definition can terminate the entire proceeding and extinguish all group members' rights. Limitation period error extinguishing class rights: AI miscalculating the interaction between the primary limitation period, the s 33ZE FCA Act suspension mechanism, and the date the representative proceeding was filed — an error of days could bar thousands of group members' claims with no recovery mechanism. PII data breach through public LLM use: a lawyer uploading a group-member registration dataset, share trading records, or discovered documents containing third-party PII to a public AI tool, triggering mandatory data breach notification to the OAIC, potential group-member tortious claims under the Privacy and Other Legislation Amendment Act 2024 statutory tort, and professional disciplinary consequences. AI-generated affidavit content filed as deponent's own evidence: AI-drafted paragraphs in a lead plaintiff's affidavit attesting to matters not within the deponent's personal knowledge — a breach of GPN-AI, NSW SC Gen 23 (where applicable), and perjury risk if the deponent swears to the truth of AI-fabricated content. - Hallucinated or misquoted judicial authority in pleadings or submissions: the post-Myer securities class action landscape, the Zonia Holdings causation ruling, and the 2021 continuous disclosure reforms mean the law changed materially in 2021-2024; AI trained on older data cites cases for superseded propositions or generates plausible-sounding non-existent decisions — directly witnessed in Australian proceedings with personal costs consequences (Murray [2025], Handa [2024], Valu [2025]). - Incorrect continuous disclosure version applied: AI applying the pre-2021 strict liability standard (s 674) to conduct occurring after the knowledge-based amendment to s 674A was enacted, or applying the temporary COVID-era regime to standard matters, producing a legally incorrect liability framework in the statement of claim that collapses at trial or summary judgment. - Group definition that fails the s 33C ascertainability test: AI-generated definitions using language importing individual merits assessment (e.g., 'group members who suffered loss') do not satisfy the requirement that membership be objectively ascertainable — a struck-out definition can terminate the entire proceeding and extinguish all group members' rights. - Limitation period error extinguishing class rights: AI miscalculating the interaction between the primary limitation period, the s 33ZE FCA Act suspension mechanism, and the date the representative proceeding was filed — an error of days could bar thousands of group members' claims with no recovery mechanism. - PII data breach through public LLM use: a lawyer uploading a group-member registration dataset, share trading records, or discovered documents containing third-party PII to a public AI tool, triggering mandatory data breach notification to the OAIC, potential group-member tortious claims under the Privacy and Other Legislation Amendment Act 2024 statutory tort, and professional disciplinary consequences. - AI-generated affidavit content filed as deponent's own evidence: AI-drafted paragraphs in a lead plaintiff's affidavit attesting to matters not within the deponent's personal knowledge — a breach of GPN-AI, NSW SC Gen 23 (where applicable), and perjury risk if the deponent swears to the truth of AI-fabricated content. ## Review checklist The lawyer's review checklist ## Review checklist Every legal authority cited in any court document has been independently verified to exist at the citation given (primary source check — not re-querying the same or any AI tool), to stand for the proposition advanced, and to be the currently authoritative version — this is the single non-negotiable check given Australian disciplinary precedents (Murray [2025] FCA 731, Handa [2024] FedCFamC2F 957, Valu [2025] FedCFamC2G 95) and the documented hallucination rates (17-33% on specialist tools; up to 91% on general LLMs). GPN-AI disclosure is included at the start of every court document where AI was used in preparation — the disclosure must identify the specific tool used, state where in the document it was applied, and describe how it was used; a blanket footer or general acknowledgment does not satisfy GPN-AI requirements; the responsible lawyer has personally verified all six statutory obligations (facts provable, authorities exist and accurate, evidence admissible, findings open, chronology accurate, discovery list conforming). Affidavits and witness statements have been drafted without AI-generated substantive content — GPN-AI requires affidavits to reflect the deponent's own recollection and knowledge; NSW SC Gen 23 imposes an absolute prohibition on AI-generated affidavit content in NSW proceedings without court leave; the deponent has read the full affidavit and confirmed each paragraph against their personal knowledge. Expert reports and expert instructions have been reviewed to confirm: (a) the expert's report reflects the expert's own reasoning and analysis, not AI-generated content; (b) instruction letters do not embed conclusions that compromise expert independence under GPN-EXPT and the Harmonised Expert Witness Code of Conduct; (c) any AI use informing materials provided to the expert is disclosed at the start of the expert report as required by GPN-AI. Limitation period calculations have been independently computed by a lawyer, accounting for s 33ZE FCA Act (suspension of limitation periods for group members from the date of filing), the applicable state Limitation Act, and any tolling or discoverability arguments — AI date calculations in multi-party class action matters carry material error risk that can extinguish all class members' claims. All group-member PII (registration data, trading data, health records, policy data) was processed only through contracted, privacy-compliant tools with confirmed data-residency and confirmed no-training-on-inputs terms — any use of public LLMs with group-member data requires an APP compliance assessment and may constitute a notifiable data breach under Part IIIC of the Privacy Act 1988 (Cth) and a serious invasion of privacy under the 2024 statutory tort. The group definition in any pleading, notice, or settlement scheme has been reviewed for legal operability (capable of objective ascertainment without individual merits assessment under s 33H FCA Act) and for consistency with the s 33C(1)(a)-(c) threshold elements — AI frequently generates definitions that look plausible but would not survive a strike-out application. The settlement distribution scheme and fee structures comply with Kain [2025] HCA 28 — no solicitors' CFO from the class fund in Federal Court proceedings; funder commissions have been reviewed against Corporations Regulations reg 7.6.01AB and ASIC RG 248 (December 2024); distribution quantum has been reviewed by an independent forensic expert against Zonia Holdings causation principles. All group-member notices and soft closure notices have been court-approved in their final form before distribution; any post-approval modification, however minor, has been the subject of a further court application; back-translation of any AI-assisted translated notice has been conducted by an accredited human translator. - Every legal authority cited in any court document has been independently verified to exist at the citation given (primary source check — not re-querying the same or any AI tool), to stand for the proposition advanced, and to be the currently authoritative version — this is the single non-negotiable check given Australian disciplinary precedents (Murray [2025] FCA 731, Handa [2024] FedCFamC2F 957, Valu [2025] FedCFamC2G 95) and the documented hallucination rates (17-33% on specialist tools; up to 91% on general LLMs). - GPN-AI disclosure is included at the start of every court document where AI was used in preparation — the disclosure must identify the specific tool used, state where in the document it was applied, and describe how it was used; a blanket footer or general acknowledgment does not satisfy GPN-AI requirements; the responsible lawyer has personally verified all six statutory obligations (facts provable, authorities exist and accurate, evidence admissible, findings open, chronology accurate, discovery list conforming). - Affidavits and witness statements have been drafted without AI-generated substantive content — GPN-AI requires affidavits to reflect the deponent's own recollection and knowledge; NSW SC Gen 23 imposes an absolute prohibition on AI-generated affidavit content in NSW proceedings without court leave; the deponent has read the full affidavit and confirmed each paragraph against their personal knowledge. - Expert reports and expert instructions have been reviewed to confirm: (a) the expert's report reflects the expert's own reasoning and analysis, not AI-generated content; (b) instruction letters do not embed conclusions that compromise expert independence under GPN-EXPT and the Harmonised Expert Witness Code of Conduct; (c) any AI use informing materials provided to the expert is disclosed at the start of the expert report as required by GPN-AI. - Limitation period calculations have been independently computed by a lawyer, accounting for s 33ZE FCA Act (suspension of limitation periods for group members from the date of filing), the applicable state Limitation Act, and any tolling or discoverability arguments — AI date calculations in multi-party class action matters carry material error risk that can extinguish all class members' claims. - All group-member PII (registration data, trading data, health records, policy data) was processed only through contracted, privacy-compliant tools with confirmed data-residency and confirmed no-training-on-inputs terms — any use of public LLMs with group-member data requires an APP compliance assessment and may constitute a notifiable data breach under Part IIIC of the Privacy Act 1988 (Cth) and a serious invasion of privacy under the 2024 statutory tort. - The group definition in any pleading, notice, or settlement scheme has been reviewed for legal operability (capable of objective ascertainment without individual merits assessment under s 33H FCA Act) and for consistency with the s 33C(1)(a)-(c) threshold elements — AI frequently generates definitions that look plausible but would not survive a strike-out application. - The settlement distribution scheme and fee structures comply with Kain [2025] HCA 28 — no solicitors' CFO from the class fund in Federal Court proceedings; funder commissions have been reviewed against Corporations Regulations reg 7.6.01AB and ASIC RG 248 (December 2024); distribution quantum has been reviewed by an independent forensic expert against Zonia Holdings causation principles. - All group-member notices and soft closure notices have been court-approved in their final form before distribution; any post-approval modification, however minor, has been the subject of a further court application; back-translation of any AI-assisted translated notice has been conducted by an accredited human translator. ## Tasks (15) | Task | Level | How AI is used | What the lawyer must review | | --- | --- | --- | --- | | [Build a viability memo: cause of action, class size and common questions](https://ailawframework.com/practice/class-actions-representative-proceedings/build-a-viability-memo-cause-of-action-class-size-and-common-questions.md) | Red | From the facts brief and sample documents, table possible claims, class-size inputs and candidate common questions, with a list of facts still needed. | Decide certification or commencement yourself Do not treat a generated common-question list as the pleaded case | | Draft the group or class definition from the claim theory | Yellow | Draft alternative class definitions (included, excluded, class period) mapped to the pleaded claims, and list over- and under-inclusion risks. | Make the definition workable for notice and administration Check it against evidence of who was actually affected | | [Run a first-pass review of the large discovery set](https://ailawframework.com/practice/class-actions-representative-proceedings/run-a-first-pass-review-of-the-large-discovery-set.md) | Yellow | From a lawyer-cleared corpus, cluster documents, flag likely hot families and draft an issues list with sample cites. | Make production and privilege calls yourself Do not load class-member personal information into a tool without a clearance decision | | Draft class notices: opt-out, registration and settlement | Red | Check the lawyer's draft notice against the court's notice orders or, before they are made, the orders sought and the court's sample notices, the class… | The lawyer writes the notice. Notice content and method are court-controlled: work from the sealed orders as made rather than a draft, confirm that the… | | Model a settlement distribution scheme from the class data | Red | From lawyer-supplied class data and the settlement heads, draft a distribution-scheme outline (eligibility, proof, tranches) and an issues list of data gaps. | Decide quantum and fairness yourself and, where required, with the court Do not invent class-member losses | | [Compare competing class proceedings and draft the carriage or consolidation note](https://ailawframework.com/practice/class-actions-representative-proceedings/compare-competing-class-proceedings-and-draft-the-carriage-or-consolidat.md) | Red | Compare competing filings on group definition, funding, lead plaintiff and proposed orders, and draft a carriage or consolidation note with document cites. | Use the current filed versions Leave carriage and lead-counsel factors to the court Re-derive every factual assertion before the note is filed | | Compile the court book and a hearing chronology | Yellow | Draft a bundle index and dated chronology from the pleaded case and key exhibits, and list missing documents. | Match the index to the final book Source every chronology row | | Draft the representative plaintiff's affidavit from the proof | Red | The tool writes no part of the affidavit — no draft, no paragraph, no heading, no jurat or annexure wording.… | The lawyer drafts the affidavit from the proof and the representative swears only their own words, checked back to what the representative actually said, with… | | Draft expert instructions and review the draft class-action report | Red | Organise the common questions, the class definition and the assumed facts the lawyer has settled into a schedule that sets the pleaded common question, the… | The lawyer writes the letter of instruction and settles every question and assumed fact in it, confirming each is supported by the pleaded common questions… | | Draft the representative pleading from the facts-and-issues brief | Red | Organise the facts-and-issues brief into a table that records beside each fact the source document and whether the brief puts the fact as shared across the… | The lawyer drafts and settles the pleading, including the group description and the common questions.… | | [Draft the settlement-approval affidavit and submissions](https://ailawframework.com/practice/class-actions-representative-proceedings/draft-the-settlement-approval-affidavit-and-submissions.md) | Red | Index the settlement deed, distribution scheme and the funding, costs and commission material as exhibits, and table every settlement sum, deduction, fee and… | Settlement approval binds the class. The lawyer writes the affidavit and the submissions, and the deponent adopts the evidence as their own knowledge.… | | Draft group-member updates and registration communications | Yellow | From the current orders and the registration process, draft member-update text and a FAQ, with a list of personal data fields that must not go into the tool. | Check court restrictions on communication with the class Do not load member lists into a public tool | | Draft a funder or client risk memo from the case file | Yellow | From the pleaded case and budget inputs, draft an internal risk memo structure (liability theory, remaining steps, budget risks) with source flags. | Do not send a generated memo to a funder Set figures and prospects yourself | | Reconcile group-member data to the class definition | Yellow | Compare the registration or client list with the class definition and produce inclusion/exclusion flags, duplicates and missing proof fields. | Decide eligibility on the class definition yourself Do not treat a generated flag as membership | | Prepare a first-pass loss-methodology issues list from sample files | Yellow | From a sample of member files and the expert's outline, list data fields needed for loss, gaps and inconsistent recording. | Leave methodology to the expert and the court Treat the list as a data-readiness check only | ## Jurisdiction notes ### Australia #### Law Part IVA Federal Court of Australia Act 1976 (Cth); Pt 10 Supreme Court Act 1986 (Vic); Pt 10 CPA (NSW); Part 9A District Court Act 1973 (NSW). #### Class-wide risk A wrong authority, group definition or limitation date can extinguish the rights of the whole class, with no individual recovery. #### Group-member data Large volumes of confidential member data. Informal AI use in review, communications and loss modelling can bypass the firm's controls. #### Court control Notices and settlements need court approval, and outcomes bind people who never instructed a lawyer. #### GPN-AI Federal Court GPN-AI (16 April 2026) imposes disclosure and verification duties. Filing unverified AI output exposes the responsible lawyer to personal adverse costs orders. ### New Zealand Local name: Representative proceedings (High Court Rules 2016) #### Procedure No dedicated class-actions statute like Part IVA of the Federal Court of Australia Act. Group claims run as representative proceedings under the High Court Rules 2016. #### Reform The Law Commission has a class-actions and litigation-funding project. It is reform work, not the current rule. #### Not Australian law Do not cite s 33C or s 33V. - Do not file a generated citation, quote or statutory reference until you have opened it in legislation.govt.nz or a trusted case database. Sources: - [High Court Rules 2016](https://www.legislation.govt.nz/regulation/public/2016/0225/104.0/whole.html) (checked 2026-09-23) - [Law Commission: class actions and litigation funding](https://www.lawcom.govt.nz/our-work/class-actions-and-litigation-funding) (checked 2026-09-23) - [Courts of New Zealand: GenAI guidelines — Lawyers (7 December 2023)](https://www.courtsofnz.govt.nz/assets/6-Going-to-Court/practice-directions/practice-guidelines/all-benches/20231207-GenAI-Guidelines-Lawyers.pdf) (checked 2026-09-23) ### United States Local name: Class actions #### Procedure Federal class actions are FRCP 23, not Part IVA. State class procedures vary. #### Filings Rule 11 and any generative-AI standing order apply to every filing, including class-certification briefs. #### Review at scale TAR is a Rule 26 proportionality and protective-order question. Keep produced material out of public tools. Sources: - [Federal Rule of Civil Procedure 11](https://www.law.cornell.edu/rules/frcp/rule_11) (checked 2026-09-23) - [Federal Rule of Civil Procedure 26](https://www.law.cornell.edu/rules/frcp/rule_26) (checked 2026-09-23) - [ABA Formal Opinion 512, Generative Artificial Intelligence Tools (29 July 2024)](https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf) (checked 2026-09-23) *Practice guidance for planning and review. Confirm current primary sources and the requirements of the actual matter. This is general information, not legal advice.*