---
title: "Defamation & Media Law"
url: "https://ailawframework.com/practice/defamation-and-media-law"
type: "practice_area"
site: "AI Law Framework"
id: "A33"
task_count: 26
last_reviewed: "2026-10-10"
---

# Defamation & Media Law

> Read the actual publication in its full context. Verify the words, audience, dates, alleged meanings and procedural requirements before advising, corresponding or publishing further material.

Canonical page: https://ailawframework.com/practice/defamation-and-media-law  
Last reviewed: 2026-10-10

Site index for agents: https://ailawframework.com/llms.txt

Also covers: Defamation, libel (historical), slander (historical), media law, reputation management, publication law, right of reply, digital intermediary liability, suppression/non-publication orders, contempt of court (sub judice and scandalising), privacy in media contexts, online safety and takedown, injurious falsehood

Defamation and media law in Australia is among the highest-stakes areas for AI-assisted legal work because liability turns on precise meaning, jurisdiction-specific procedural preconditions (concerns notices, serious harm threshold), a non-uniform and still-rolling reform landscape (Stage 2 enacted in NSW/ACT from 1 July 2024, Vic from 11 September 2024, Tas from 2025, Qld enacted 11 December 2025 with commencement seven days after Royal Assent, SA partial only as at mid-2026), and factual nuance that public LLMs hallucinate at 17-33% rates (Magesh et al.) rising above 58% on some query types (Dahl et al.). The combination of a one-year limitation period running from first publication (not discovery), a mandatory concerns-notice precondition, rapidly evolving statutory defences varying by jurisdiction, suppression-order contempt risks, and the area's own emergence as a subject of AI-defamation litigation makes unchecked AI output a direct client-harm risk with limited opportunity for correction once deadlines pass.

## Key facts

- Tasks: 26
- Green: 4, Yellow: 17, Red: 5
- 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.
- **Reconcile the figures**: Trace every relied-on figure to its source, confirm dates and assumptions, and independently recalculate totals or derived amounts.

## Checklist

- Open each relied-on authority independently. Confirm its existence, current status, proposition and any quotation or pinpoint reference.
- 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.
- Confirm that the wording reflects the client’s instructions and contains no unintended admission, concession or commitment. Check the recipient and tone.
- 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: Defamation, libel (historical), slander (historical), media law, reputation management, publication law, right of reply, digital intermediary liability, suppression/non-publication orders, contempt of court (sub judice and scandalising), privacy in media contexts, online safety and takedown, injurious falsehood

Defamation and media law in Australia is among the highest-stakes areas for AI-assisted legal work because liability turns on precise meaning, jurisdiction-specific procedural preconditions (concerns notices, serious harm threshold), a non-uniform and still-rolling reform landscape (Stage 2 enacted in NSW/ACT from 1 July 2024, Vic from 11 September 2024, Tas from 2025, Qld enacted 11 December 2025 with commencement seven days after Royal Assent, SA partial only as at mid-2026), and factual nuance that public LLMs hallucinate at 17-33% rates (Magesh et al.) rising above 58% on some query types (Dahl et al.). The combination of a one-year limitation period running from first publication (not discovery), a mandatory concerns-notice precondition, rapidly evolving statutory defences varying by jurisdiction, suppression-order contempt risks, and the area's own emergence as a subject of AI-defamation litigation makes unchecked AI output a direct client-harm risk with limited opportunity for correction once deadlines pass.

## Limits on AI assistance

Red lines — never in this area

## Limits on AI assistance

AI must not determine, state, or confirm any limitation period date — all limitation calculations must be performed by the responsible lawyer from verified first-publication date evidence. The one-year period runs from first publication (not discovery) and continues during the concerns-notice period. AI must not draft, finalise, or certify a concerns notice without a full lawyer review of every statutory element, including: correct identification of the publisher-respondent, particulars of each imputation, and factually specific (not formulaic) description of serious harm. A facially defective notice bars the claim as a matter of substantive law (WA Supreme Court 2024). AI must not derive or finalise defamatory imputations — the ordinary reasonable reader test is a legal judgment requiring the lawyer to read the actual publication and apply the test; AI inference from training data does not replicate this. AI must not be used to process, summarise, or analyse any material subject to a suppression or non-publication order — doing so risks contempt of court and discloses suppressed content to a third-party server. No AI-produced case citation may be used in any court filing, client advice, correspondence, or settlement negotiation without independent verification on a primary legal database (AustLII, LexisNexis AU, Westlaw AU) — citation hallucination is confirmed by Australian court incident reports including Dayal [2024] FedCFamC2F 1166 (solicitor sanctioned by Victorian Legal Services Board) and an empirical hallucination rate of 17-33% in specialised legal AI tools. AI must not advise on which jurisdiction's version of the Defamation Act applies or whether a specific Stage 2 provision is operative without a lawyer independently confirming commencement — Stage 2 enactment dates differ by state and the position is still evolving as at mid-2026 (SA digital intermediary provisions absent; Qld enacted December 2025, commencement date to be confirmed; Tas Defamation Amendment Act 2025 commencement to be confirmed). AI must not independently assess whether the serious harm threshold is met — that determination requires primary evidence about the actual extent of publication, the identity and characteristics of recipients, and what recipients understood, none of which AI models possess. AI must not generate content that could itself constitute defamation of a real identifiable person, whether in example imputations, hypothetical scenarios, corrective statements, or training materials. AI-as-publisher liability is unresolved in Australia but the risk exists. AI must not state the non-economic loss cap figure as the ceiling for all defamation damages — aggravated damages are not capped and may significantly exceed it; and the cap is jurisdiction-specific and annually indexed, so any AI-stated figure must be independently verified before it is communicated to a client or court. AI must not substitute for the final go/no-go publication decision in pre-publication review — that judgment belongs to the lawyer and must be supported by the lawyer's independent analysis of meaning, defences and evidence.

- AI must not determine, state, or confirm any limitation period date — all limitation calculations must be performed by the responsible lawyer from verified first-publication date evidence. The one-year period runs from first publication (not discovery) and continues during the concerns-notice period.
- AI must not draft, finalise, or certify a concerns notice without a full lawyer review of every statutory element, including: correct identification of the publisher-respondent, particulars of each imputation, and factually specific (not formulaic) description of serious harm. A facially defective notice bars the claim as a matter of substantive law (WA Supreme Court 2024).
- AI must not derive or finalise defamatory imputations — the ordinary reasonable reader test is a legal judgment requiring the lawyer to read the actual publication and apply the test; AI inference from training data does not replicate this.
- AI must not be used to process, summarise, or analyse any material subject to a suppression or non-publication order — doing so risks contempt of court and discloses suppressed content to a third-party server.
- No AI-produced case citation may be used in any court filing, client advice, correspondence, or settlement negotiation without independent verification on a primary legal database (AustLII, LexisNexis AU, Westlaw AU) — citation hallucination is confirmed by Australian court incident reports including Dayal [2024] FedCFamC2F 1166 (solicitor sanctioned by Victorian Legal Services Board) and an empirical hallucination rate of 17-33% in specialised legal AI tools.
- AI must not advise on which jurisdiction's version of the Defamation Act applies or whether a specific Stage 2 provision is operative without a lawyer independently confirming commencement — Stage 2 enactment dates differ by state and the position is still evolving as at mid-2026 (SA digital intermediary provisions absent; Qld enacted December 2025, commencement date to be confirmed; Tas Defamation Amendment Act 2025 commencement to be confirmed).
- AI must not independently assess whether the serious harm threshold is met — that determination requires primary evidence about the actual extent of publication, the identity and characteristics of recipients, and what recipients understood, none of which AI models possess.
- AI must not generate content that could itself constitute defamation of a real identifiable person, whether in example imputations, hypothetical scenarios, corrective statements, or training materials. AI-as-publisher liability is unresolved in Australia but the risk exists.
- AI must not state the non-economic loss cap figure as the ceiling for all defamation damages — aggravated damages are not capped and may significantly exceed it; and the cap is jurisdiction-specific and annually indexed, so any AI-stated figure must be independently verified before it is communicated to a client or court.
- AI must not substitute for the final go/no-go publication decision in pre-publication review — that judgment belongs to the lawyer and must be supported by the lawyer's independent analysis of meaning, defences and evidence.

## Jurisdiction-specific considerations

Australian traps

## Jurisdiction-specific considerations

Stage 2 defamation reforms are operative as follows as at mid-2026: NSW and ACT from 1 July 2024; Victoria from 11 September 2024; Tasmania — Defamation Amendment Act 2025 enacted (commencement date to be independently confirmed); Queensland — Defamation and Other Legislation Amendment Bill 2025 passed Parliament 11 December 2025, commenced seven days after Royal Assent (date of Royal Assent must be independently confirmed before advising); South Australia — partial reforms only, digital intermediary innocent dissemination defence absent. AI tools trained before or during this roll-out will state incorrect current law depending on jurisdiction; always verify applicable statute version. The concerns notice (s 12A/12B uniform Acts, as amended) is a substantive, not merely procedural, precondition — failure invalidates the claim and is not curable by amendment if the limitation period has expired. The WA Supreme Court confirmed the substantive nature in 2024. AI drafts commonly produce boilerplate 'serious harm' language that Australian courts have explicitly rejected (Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991). The one-year limitation period runs from the date of first publication (not discovery) and continues to run during the concerns-notice period — serving a notice does not stop the clock. AI routinely imports discovery-based limitation rules from personal injury or contract, and extension applications are consistently refused where the plaintiff simply failed to act promptly (Richards v Hutchinson [2024] QSC 73). The serious harm threshold (s 10A): must be established on actual facts specific to the plaintiff's reputation and the reception of the publication. Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991 rejected a formulaic approach. AI drafts frequently reproduce the rejected boilerplate formula. The non-economic loss cap under s 35 is indexed annually and differs between states — as at 2025-2026 it is approximately $478,500 in NSW and approximately $500,000 in Queensland and South Australia; other jurisdictions have different indexed amounts. Aggravated damages are not subject to the cap. AI-stated figures are unreliable and must be verified in the applicable jurisdiction's current gazette or legislation before any advice or submission. Corporate plaintiffs cannot sue in defamation for general damages — only economic loss — and only if the corporation has fewer than 10 employees and is not a public body (s 9 uniform Acts). This is a common AI error that exposes clients to struck-out claims. The digital intermediary innocent dissemination defence introduced by Stage 2 is not available in South Australia and was only enacted in Queensland in late 2025 (commencement to be confirmed). The defence requires an accessible complaints mechanism and removal or blocking within seven days of a valid complaint, and is unavailable where the intermediary promoted or participated in the material. Staged enactment means AI will systematically overstate availability. Suppression and non-publication order regimes differ across courts and jurisdictions: the Court Suppression and Non-publication Orders Act 2010 (NSW) governs NSW Supreme Court proceedings; separate provisions govern federal courts; Victoria, Queensland, SA and WA each have their own regimes. Uploading suppressed material to any cloud-hosted AI tool is a potential contempt of court and a data-exfiltration risk. The single publication rule (introduced Stage 1, 2021) means time runs from when a publication is first made publicly available, not each subsequent download or view. Online content can present genuinely contested application of this rule — AI may apply the older multiple-publication rule drawn from common law or foreign jurisdictions. Search engine liability: Google LLC v Defteros [2022] HCA 27 confirmed that organic search returns hyperlinking to defamatory third-party content are generally not a publication by the search engine. Stage 2 codifies this exemption but excludes sponsored results. AI may cite Defteros incorrectly or fail to distinguish organic from sponsored search results. Sub judice contempt is a separate and acute risk in media defamation matters that is not covered by the defamation Act framework — a lawyer advising a media client on pre-publication review must independently check whether active proceedings create sub judice constraints, particularly for criminal trials. The VLRC 2024 report recommended reforms to the contempt framework; current law must be verified before advising. Inputting client-identifying information, draft publications, or suppressed material into public generative AI tools engages potential breach of ASCR r 9.1 (confidentiality). The Dayal matter [2024] FedCFamC2F 1166 is the leading Australian disciplinary precedent for unsupervised AI-generated court content; the solicitor was prohibited from handling trust money or practising unsupervised for two years. Ignorance of hallucination risk is no defence. AI-generated defamatory content is an emerging head of liability. The Brian Hood v OpenAI matter (the first known Australian AI-defamation action) was abandoned by Hood in February 2024 after OpenAI corrected ChatGPT; no merits judgment was entered. The framework for AI-as-publisher under Australian law is unresolved — do not advise clients on it without a current primary-source review.

- Stage 2 defamation reforms are operative as follows as at mid-2026: NSW and ACT from 1 July 2024; Victoria from 11 September 2024; Tasmania — Defamation Amendment Act 2025 enacted (commencement date to be independently confirmed); Queensland — Defamation and Other Legislation Amendment Bill 2025 passed Parliament 11 December 2025, commenced seven days after Royal Assent (date of Royal Assent must be independently confirmed before advising); South Australia — partial reforms only, digital intermediary innocent dissemination defence absent. AI tools trained before or during this roll-out will state incorrect current law depending on jurisdiction; always verify applicable statute version.
- The concerns notice (s 12A/12B uniform Acts, as amended) is a substantive, not merely procedural, precondition — failure invalidates the claim and is not curable by amendment if the limitation period has expired. The WA Supreme Court confirmed the substantive nature in 2024. AI drafts commonly produce boilerplate 'serious harm' language that Australian courts have explicitly rejected (Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991).
- The one-year limitation period runs from the date of first publication (not discovery) and continues to run during the concerns-notice period — serving a notice does not stop the clock. AI routinely imports discovery-based limitation rules from personal injury or contract, and extension applications are consistently refused where the plaintiff simply failed to act promptly (Richards v Hutchinson [2024] QSC 73).
- The serious harm threshold (s 10A): must be established on actual facts specific to the plaintiff's reputation and the reception of the publication. Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991 rejected a formulaic approach. AI drafts frequently reproduce the rejected boilerplate formula.
- The non-economic loss cap under s 35 is indexed annually and differs between states — as at 2025-2026 it is approximately $478,500 in NSW and approximately $500,000 in Queensland and South Australia; other jurisdictions have different indexed amounts. Aggravated damages are not subject to the cap. AI-stated figures are unreliable and must be verified in the applicable jurisdiction's current gazette or legislation before any advice or submission.
- Corporate plaintiffs cannot sue in defamation for general damages — only economic loss — and only if the corporation has fewer than 10 employees and is not a public body (s 9 uniform Acts). This is a common AI error that exposes clients to struck-out claims.
- The digital intermediary innocent dissemination defence introduced by Stage 2 is not available in South Australia and was only enacted in Queensland in late 2025 (commencement to be confirmed). The defence requires an accessible complaints mechanism and removal or blocking within seven days of a valid complaint, and is unavailable where the intermediary promoted or participated in the material. Staged enactment means AI will systematically overstate availability.
- Suppression and non-publication order regimes differ across courts and jurisdictions: the Court Suppression and Non-publication Orders Act 2010 (NSW) governs NSW Supreme Court proceedings; separate provisions govern federal courts; Victoria, Queensland, SA and WA each have their own regimes. Uploading suppressed material to any cloud-hosted AI tool is a potential contempt of court and a data-exfiltration risk.
- The single publication rule (introduced Stage 1, 2021) means time runs from when a publication is first made publicly available, not each subsequent download or view. Online content can present genuinely contested application of this rule — AI may apply the older multiple-publication rule drawn from common law or foreign jurisdictions.
- Search engine liability: Google LLC v Defteros [2022] HCA 27 confirmed that organic search returns hyperlinking to defamatory third-party content are generally not a publication by the search engine. Stage 2 codifies this exemption but excludes sponsored results. AI may cite Defteros incorrectly or fail to distinguish organic from sponsored search results.
- Sub judice contempt is a separate and acute risk in media defamation matters that is not covered by the defamation Act framework — a lawyer advising a media client on pre-publication review must independently check whether active proceedings create sub judice constraints, particularly for criminal trials. The VLRC 2024 report recommended reforms to the contempt framework; current law must be verified before advising.
- Inputting client-identifying information, draft publications, or suppressed material into public generative AI tools engages potential breach of ASCR r 9.1 (confidentiality). The Dayal matter [2024] FedCFamC2F 1166 is the leading Australian disciplinary precedent for unsupervised AI-generated court content; the solicitor was prohibited from handling trust money or practising unsupervised for two years. Ignorance of hallucination risk is no defence.
- AI-generated defamatory content is an emerging head of liability. The Brian Hood v OpenAI matter (the first known Australian AI-defamation action) was abandoned by Hood in February 2024 after OpenAI corrected ChatGPT; no merits judgment was entered. The framework for AI-as-publisher under Australian law is unresolved — do not advise clients on it without a current primary-source review.

## Risks to address

How it goes wrong

## Risks to address

AI fabricates defamation case citations — confirmed by Australian court incident reports (Dayal [2024] FedCFamC2F 1166; solicitor sanctioned) and empirical research showing 17-33% hallucination rates in specialised legal AI tools. Defamation practitioners cite cases on imputation meaning, serious harm, defence elements and quantum at high frequency — each citation is a hallucination risk. AI applies the wrong jurisdiction's defamation statute — the uniform Acts are not uniform. AI trained on pre-Stage-2 material, or generalising across all jurisdictions, states law that is not operative in the applicable state (e.g., treating SA as having the digital intermediary defence, or treating Queensland as fully Stage-2 compliant before confirmed commencement). AI produces a formulaic concerns notice that is substantively invalid — courts have rejected boilerplate serious harm language (Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991). A defective notice filed without lawyer review may extinguish the client's claim if held invalid and the limitation period has expired with no viable extension argument. AI miscalculates the limitation deadline — importing discovery-based rules from tort or contract, applying the wrong start date, failing to apply the single publication rule correctly for online content, or treating the concerns-notice period as suspending the limitation clock (it does not). AI conflates the ordinary reasonable reader test with literal, pedantic, or LLM-trained-corpus meaning — producing imputations that are too narrow (missing the sting) or too broad (unsustainable on the words), damaging the client's position in ways that may not be correctable after limitation. AI states the non-economic loss cap incorrectly — using an outdated or wrong-jurisdiction figure, or treating it as the ceiling for all defamation damages when aggravated damages are uncapped and have produced awards significantly exceeding the s 35 cap in Australian courts.

- AI fabricates defamation case citations — confirmed by Australian court incident reports (Dayal [2024] FedCFamC2F 1166; solicitor sanctioned) and empirical research showing 17-33% hallucination rates in specialised legal AI tools. Defamation practitioners cite cases on imputation meaning, serious harm, defence elements and quantum at high frequency — each citation is a hallucination risk.
- AI applies the wrong jurisdiction's defamation statute — the uniform Acts are not uniform. AI trained on pre-Stage-2 material, or generalising across all jurisdictions, states law that is not operative in the applicable state (e.g., treating SA as having the digital intermediary defence, or treating Queensland as fully Stage-2 compliant before confirmed commencement).
- AI produces a formulaic concerns notice that is substantively invalid — courts have rejected boilerplate serious harm language (Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991). A defective notice filed without lawyer review may extinguish the client's claim if held invalid and the limitation period has expired with no viable extension argument.
- AI miscalculates the limitation deadline — importing discovery-based rules from tort or contract, applying the wrong start date, failing to apply the single publication rule correctly for online content, or treating the concerns-notice period as suspending the limitation clock (it does not).
- AI conflates the ordinary reasonable reader test with literal, pedantic, or LLM-trained-corpus meaning — producing imputations that are too narrow (missing the sting) or too broad (unsustainable on the words), damaging the client's position in ways that may not be correctable after limitation.
- AI states the non-economic loss cap incorrectly — using an outdated or wrong-jurisdiction figure, or treating it as the ceiling for all defamation damages when aggravated damages are uncapped and have produced awards significantly exceeding the s 35 cap in Australian courts.

## Review checklist

The lawyer's review checklist

## Review checklist

Every case citation in AI output has been independently verified on AustLII, LexisNexis AU, or Westlaw AU — confirmed to exist, to be on point for the proposition asserted, and not to have been overruled, distinguished, or superseded by Stage 2 amendments in ways the AI did not identify. The jurisdiction's current version of the Defamation Act has been confirmed, including: whether Stage 2 Part A (digital intermediary defence) and Part B (police-reporting immunity) are operative in the relevant state or territory, and the correct commencement date (NSW/ACT 1 July 2024; Vic 11 September 2024; Tas 2025; Qld late 2025 — Royal Assent date to be confirmed; SA partial only). Every imputation was derived by the reviewing lawyer from the actual words and context of the publication — not accepted from AI output — and each is capable of being understood by the ordinary reasonable reader in the way pleaded. The AI's imputation list has been treated as a prompt for the lawyer's own analysis, not as the analysis itself. The concerns notice (if AI-assisted) satisfies every element of s 12A/12B: correct respondent identified, full particulars of each imputation relied on, and a factually specific description of serious harm grounded in actual evidence of impact — not boilerplate formula rejected in Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991. All limitation dates and procedural deadlines have been calculated by the lawyer from verified first-publication dates on primary evidence; the concerns-notice clock interaction has been checked; AI output on dates has not been relied on without independent primary-source confirmation. The non-economic loss cap figure stated in any advice, submission, or settlement communication is the current indexed amount for the applicable jurisdiction at the applicable date — independently verified from the jurisdiction's current gazette or legislation, not from AI output. No suppressed, non-publication-ordered, or otherwise court-restricted material was input into any AI tool; and no client-identifying information, matter strategy, or draft publication content was processed in a public (non-firm-approved) generative AI tool, consistent with ASCR r 9.1. Where AI output addresses frontier legal issues — AI-generated defamation, digital intermediary liability in non-Stage-2 jurisdictions, the search engine sponsored/organic distinction, or sub judice contempt — the lawyer has confirmed the current Australian legal position from a primary source before any advice or filing is based on it.

- Every case citation in AI output has been independently verified on AustLII, LexisNexis AU, or Westlaw AU — confirmed to exist, to be on point for the proposition asserted, and not to have been overruled, distinguished, or superseded by Stage 2 amendments in ways the AI did not identify.
- The jurisdiction's current version of the Defamation Act has been confirmed, including: whether Stage 2 Part A (digital intermediary defence) and Part B (police-reporting immunity) are operative in the relevant state or territory, and the correct commencement date (NSW/ACT 1 July 2024; Vic 11 September 2024; Tas 2025; Qld late 2025 — Royal Assent date to be confirmed; SA partial only).
- Every imputation was derived by the reviewing lawyer from the actual words and context of the publication — not accepted from AI output — and each is capable of being understood by the ordinary reasonable reader in the way pleaded. The AI's imputation list has been treated as a prompt for the lawyer's own analysis, not as the analysis itself.
- The concerns notice (if AI-assisted) satisfies every element of s 12A/12B: correct respondent identified, full particulars of each imputation relied on, and a factually specific description of serious harm grounded in actual evidence of impact — not boilerplate formula rejected in Russell v S3@Raw Pty Ltd (No 3) [2024] FCA 991.
- All limitation dates and procedural deadlines have been calculated by the lawyer from verified first-publication dates on primary evidence; the concerns-notice clock interaction has been checked; AI output on dates has not been relied on without independent primary-source confirmation.
- The non-economic loss cap figure stated in any advice, submission, or settlement communication is the current indexed amount for the applicable jurisdiction at the applicable date — independently verified from the jurisdiction's current gazette or legislation, not from AI output.
- No suppressed, non-publication-ordered, or otherwise court-restricted material was input into any AI tool; and no client-identifying information, matter strategy, or draft publication content was processed in a public (non-firm-approved) generative AI tool, consistent with ASCR r 9.1.
- Where AI output addresses frontier legal issues — AI-generated defamation, digital intermediary liability in non-Stage-2 jurisdictions, the search engine sponsored/organic distinction, or sub judice contempt — the lawyer has confirmed the current Australian legal position from a primary source before any advice or filing is based on it.

## Tasks (26)

| Task | Level | How AI is used | What the lawyer must review |
| --- | --- | --- | --- |
| List candidate imputations from the publication | Red | From the article, post or broadcast, list candidate meanings an ordinary reader might take, with the words that support each meaning. | Derive each imputation from the raw publication; they are the claim Do not plead a generated meaning you have not adopted |
| Draft a pre-action concerns or retraction notice | Red | Extract the publication particulars from the saved copy of the matter — publisher, date, where the matter can still be accessed, and whether it is still up,… | The lawyer writes the notice and owns the imputations and the harm it alleges, because what the notice specifies fixes what can later be pleaded.… |
| [Build a serious-harm or threshold-evidence table](https://ailawframework.com/practice/defamation-and-media-law/build-a-serious-harm-or-threshold-evidence-table.md) | Yellow | From the analytics, correspondence and witness notes, list evidence of extent, gravity and actual impact, with gaps marked. | Apply the local threshold test; it differs by country Do not leave a generated 'serious harm' label in advice |
| [Build a defences checklist from the publication and the file](https://ailawframework.com/practice/defamation-and-media-law/build-a-defences-checklist-from-the-publication-and-the-file.md) | Yellow | List truth, honest opinion, privilege, public-interest and contextual-truth (or local equivalents) against the facts in the file, with missing evidence on… | Confirm which defences the governing Act actually contains Do not assume Stage 2 digital defences apply everywhere |
| Map platform or intermediary liability from the publication path | Yellow | From the URL, host and complaint history, list who published, who hosted and which complaints-procedure or innocent-dissemination facts are in the file. | Confirm the current intermediary statute before you write Remember the rules differ by country and are still moving |
| Calculate limitation and first-publication dates from the capture | Red | From the capture files, list first-publication dates, later downloads and any single-publication or multiple-publication facts into a date table. | Check whether limitation runs from first publication, not discovery; it can be one year Confirm the rule and start date that apply |
| Draft a takedown or complaint pack for AI-generated defamatory content | Yellow | Produce a first-draft platform complaint or letter from the generated output, the prompts if known and the remedy the lawyer has named. | Decide who the publisher is Do not assume a chatbot operator is liable because a draft says so |
| Draft a statement of claim, defence or particulars of publication | Red | The tool extracts from the publication capture the exact words complained of with their date, headline, byline, placement and any share or download figures,… | The lawyer settles every imputation and particular in their own words and satisfies themselves that each imputation is reasonably capable of arising from the… |
| Draft suppression or non-publication submissions from the harm evidence | Yellow | Produce a first-draft submissions outline from the open-justice test the lawyer has named and the harm evidence on the file. | Start from open justice as the default Confirm the statute that actually applies to this court |
| Draft a settlement deed and a corrective statement or apology | Yellow | Produce a first-draft deed, offer letter and corrective text from the agreed points, marking wording that could be a fresh publication. | Check the apology does not defame someone else Settle every word before it is published |
| Run a pre-publication defamation review of the draft copy | Yellow | Flag statements that may defame identified people, list candidate meanings and note where the file has no support. | Identify every person the copy is about, including unnamed but identifiable people Remember sign-off creates publication risk |
| Build a sub judice risk table for a draft publication | Yellow | From the draft and the current proceedings list, flag passages that identify parties, alleged facts or likely witnesses in active cases. | Check contempt risk around criminal, coronial and jury matters Do not treat the table as clearance |
| Build a comparable-awards table for damages discussions | Yellow | From the authorities the lawyer has collected, list awards by injury type, with citations and a note of caps that may apply. | Open every cited judgment Check caps; they are indexed and jurisdiction-specific |
| Build an injurious-falsehood or malicious-falsehood issues table | Yellow | From the publication and the trading-loss papers, list falsity, malice and special-damage facts with document references. | Treat this as a different cause of action Do not treat a defamation defence as an answer to it |
| Capture the publication with URL, time, audience and copies | Green | Build a capture record from the URL, screenshots, archive copies and analytics exports, listing what is still missing. | Confirm the capture is complete and dated Do not plead a later edit of a live page as the publication |
| Compare the published text with a proposed correction | Green | Compare the original capture with the draft correction and list meanings that remain, meanings withdrawn and any new sting. | Read both texts as a first-time reader would Check the correction does not make things worse |
| Draft a without-prejudice offer of amends from the notice and the capture | Yellow | Produce a first-draft offer covering correction, apology, take-down and costs from the notice and the lawyer's settlement range. | Check the offer-of-amends rules; they are technical Get timing and form right; they can be a defence or a trap |
| [Draft an eSafety removal-notice request for AI-generated intimate or defamatory imagery](https://ailawframework.com/practice/defamation-and-media-law/draft-an-esafety-removal-notice-request-for-ai-generated-intimate-or-def.md) | Yellow | Assemble the URL capture, harm summary and complaint text from the lawyer-selected material. | Verify capture metadata; do not upload the imagery itself to any AI tool. |
| Assess a client's complaint about AI chatbot outputs defaming them | Yellow | Collect and log the outputs and publication facts. | Lawyer assesses publication and defences. |
| Draft a concerns notice about AI-generated defamatory content | Red | Draft a notice from the logged outputs. | Lawyer settles notice under the uniform Act. |
| Review publication for AI-generated images or deepfakes | Yellow | Flag likely synthetic content and identification risks. | Lawyer assesses legal exposure. |
| Check submissions for hallucinated defamation authorities | Green | Verify every citation exists. | Lawyer confirms in AustLII/JADE. |
| Summarise an online-harm takedown pathway | Yellow | Map eSafety and platform removal options. | Lawyer confirms current law. |
| Advise on AI platform liability as a publisher | Yellow | Apply publication law to the AI platform. | Lawyer advises. |
| Assess Stage 2 defamation reforms for an AI chatbot output | Yellow | Analyse. | Lawyer advises. |
| Explain to a client the risks of AI-assisted defamation litigation | Green | Draft a warning. | Lawyer approves. |

## Jurisdiction notes

### New Zealand

#### Governing Act

Defamation Act 1992, not the uniform Australian Defamation Acts.

#### Not Australian law

Do not apply the serious-harm threshold, concerns notices or Stage 2 digital-intermediary defences without confirming the same rule in the New Zealand Act.

#### Deepfakes

The Courts of New Zealand guidelines warn that generated multimedia can be fake and raise authenticity issues in evidence.

Sources:

- [Defamation Act 1992](https://www.legislation.govt.nz/act/public/1992/105/en/latest/) (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: Defamation and media law

#### State law

State law governs this work. Check the statutes, court rules and conduct rules of the state where the matter is. Do not import an Australian statute, form or court AI practice note.

#### Defamation

State tort law with a First Amendment overlay. No uniform national Defamation Act.

Sources:

- [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)
- [Federal Rule of Civil Procedure 11](https://www.law.cornell.edu/rules/frcp/rule_11) (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.*