The AI Law Report Issue 003 | Friday 21 August 2026
Global developments from the courts, regulators and public authorities
Issue 003 | Friday 21 August 2026
Reporting period: 14–21 August 2026
Published by AI Policies UK
IN THIS ISSUE
This edition contains 29 entries: 24 developments reported for the first time, 4 updates to matters opened in earlier issues, and 1 correction to Issue 002. A further 32 matters were captured during the reporting period but fall below the publication threshold or are held pending primary material; they are retained in the underlying record and identified in the Methodology note.
As in previous editions, the Report also includes material developments identified through the seven-day backfill where the underlying event predates the formal reporting period but had not previously entered the record.
The 29 entries arise in 11 national jurisdictions, Australia, Brazil, India, Ireland, Italy, the Philippines, Thailand, Türkiye, the United Kingdom, the United States and Vietnam. Jurisdictions are counted once at national level: federal, state and local layers within a single country are not counted separately, so New Mexico, Colorado, California and the federal courts all count once under the United States.
Matter numbers (ALR-) are permanent references into the Report's underlying record. A matter retains its number for as long as it remains live, so that its history can be followed across editions.
Correction to Issue 002
ALR-0033 Brazil: the TSE internal meeting reported for 13 August did not take place
Lead Developments
ALR-0070 Vietnam: national list of high-risk AI systems enters into force
ALR-0071 United States: core privacy claims survive dismissal in In re Otter.AI Privacy Litigation
ALR-0072 United States: labels permitted to add DMCA anti-circumvention claim in UMG v Suno
ALR-0039 United States: multistate youth-safety trial against Meta reaches opening statements UPDATE
ALR-0073 Brazil: TSE orders removal of AI-generated electoral video in a new representation
ALR-0074 United States: Babylon Bee challenges New Mexico's AI political-media law
Courts & Litigation
ALR-0075 United States: Round Hill Music sues Anthropic over lyrics used in training NEW
ALR-0076 United States: Round Hill Music sues Suno over compositions used in training NEW
ALR-0080 United States: discovery sanctions contested in the OpenAI copyright MDL NEW
ALR-0081 United States: Zuckerberg deposition permitted in coordinated Meta AI copyright suits NEW
ALR-0082 United States: Apple shareholder derivative action alleges failure to oversee AI legal risk NEW
ALR-0077 United States: Grok privacy class action stays in California NEW
ALR-0084 United States: Court of Federal Claims rejects Scale AI's bid protest NEW
ALR-0009 United States: Minnesota opposes xAI's injunction application; hearing held UPDATE
ALR-0078 India: Delhi High Court grants interim relief to Khushi Kapoor NEW
ALR-0079 India: Chandigarh court orders 72-hour takedown of alleged AI-manipulated political video NEW
ALR-0085 / ALR-0086 Brazil: two São Paulo civil actions over AI-generated political videos NEW
ALR-0083 Philippines: AI-generated demonstrative material used in the Duterte impeachment trial NEW
Regulators, Enforcement & Oversight
ALR-0087 England and Wales: Charity Commission places AI within the sector risk environment NEW
ALR-0051 United States: NAIC AI Risk Evaluation Supplement pilot confirmed by official summary UPDATE
ALR-0052 Brazil: ANPD generative-AI session takes place UPDATE
Government & Institutional Developments
ALR-0088 Türkiye: AI Action Plan 2026–2030 moves into implementation by Presidential Circular NEW
ALR-0089 New South Wales: Office of AI to be established within the Cabinet Office NEW
ALR-0090 Ireland: consultation opens on the EU Cloud and AI Development Act NEW
Legislation & Rulemaking
ALR-0091 Colorado: Attorney General files proposed automated decision-making and chatbot rules NEW
ALR-0092 Thailand: consultation closes on the revised draft AI Act NEW
ALR-0093 Italy: parliamentary-opinion deadline passes on the EU AI Act implementing decree NEW
Developments to Watch
Twenty-four live matters awaiting their next trigger, listed at the back of this edition.
THIS WEEK
The Friday-to-Friday record for 14–21 August covers Australia, Brazil, India, Ireland, Italy, the Philippines, Thailand, Türkiye, the United Kingdom, the United States and Vietnam.
The clearest single development is regulatory rather than judicial. On 15 August Vietnam's list of high-risk artificial-intelligence systems took effect, identifying particular uses in healthcare, education, finance, judicial and legal proceedings, biometrics and transport as high risk, requiring human supervision and stating expressly that the use of AI does not displace the legal responsibility of the competent person or body. That is operative national AI regulation, not a consultation or a strategy document, and only a small number of jurisdictions have yet reached that stage.
In the courts, the week's material clusters around what happened before an AI system produced anything. In Massachusetts, the record companies suing Suno were permitted to add a claim under the anti-circumvention provisions of the Digital Millennium Copyright Act, on the theory that training material was obtained by stream-ripping recordings from YouTube. In California, Round Hill Music filed two new actions on the same day, against Anthropic and against Suno, each concerned with the provenance of training inputs. In the Southern District of New York, the publishers' sanctions application against OpenAI turned on preservation, searchability and a disputed sample of roughly 20 million ChatGPT conversations. In the Northern District of California, a shareholder derivative action against Apple's directors placed the same subject matter — how training material was acquired into the language of board oversight.
Alongside that, conventional privacy law was applied to an AI assistant. Judge Eumi K. Lee allowed federal and state wiretap claims, Illinois biometric claims and unfair-competition claims to proceed in the consolidated Otter.AI litigation, which concerns an AI notetaker said to join meetings, record and transcribe participants who are not themselves users, and process the resulting material. Nothing has been decided about whether Otter did anything unlawful. What has been decided is that statutes written for interception and biometric identifiers are capable of reaching the conduct as pleaded.
Synthetic political media produced developments in three jurisdictions on very different legal footings. Brazil's Superior Electoral Court granted interim removal, non-republication and disclosure relief in a fresh representation concerning an AI-generated video depicting a presidential candidate. Two civil actions filed in São Paulo on the same day produced opposite interim results, one still pending and one refused on the basis that the material appeared to fall within satire and political criticism. In New Mexico, a satirical publisher filed a federal First Amendment challenge to the state's disclosure requirements for AI-manipulated political communications. In Delhi and Chandigarh, Indian courts continued to grant interim takedown relief over synthetic depictions of identifiable individuals.
On the public-authority side, Türkiye's national AI Action Plan moved into implementation through a Presidential Circular; New South Wales announced an Office of AI within its Cabinet Office; Ireland opened consultation on the European Commission's proposed Cloud and AI Development Act; Colorado's Attorney General filed proposed rules implementing the state's automated decision-making and chatbot statutes; Thailand's consultation on a revised draft AI Act closed; and the Charity Commission for England and Wales identified AI within the risk environment for charities while reminding trustees that safeguards and oversight remain their responsibility.
One matter has been corrected. Issue 002 reported that ministers of Brazil's Superior Electoral Court met on 13 August to align their approach to the electoral deepfake rules. That meeting did not take place.
CORRECTION TO ISSUE 002 — BRAZIL, TSE
ALR-0033 · Correction
Issue 002 recorded, on the basis of contemporaneous Brazilian reporting and with partial verification expressly stated, that a closed meeting of the TSE's ministers had been convened on 13 August 2026 in the Flávio Bolsonaro AI/deepfake proceeding (Federação Brasil da Esperança v Partido Liberal and Flávio Nantes Bolsonaro, Rp 0601315-97.2026.6.00.0000) to align the court's approach before formal judgment.
Subsequent verification establishes that the meeting did not take place. It had been intended to follow the court's morning plenary session; the plenary overran and the meeting was postponed. Reporting citing the TSE's press office states that no replacement date had then been fixed.
Accordingly, there was no collective interpretation of the electoral AI rules, no consensus among the ministers and no merits determination on 13 August. The record for that date now reads: scheduled internal TSE meeting concerning AI and deepfake electoral rules postponed because the plenary session overran; no merits determination or collective interpretation reached.
The underlying representation, the 10 August defence submission and the substantive legal question recorded in Issue 002 are unaffected. The proceeding remains pending. Some reporting suggested a possible rescheduled meeting on 18 August; that has not been confirmed from any court source and is not treated as a fixed procedural date.
What the record indicates
Four observations on the week's record. They are offered as observations, not predictions.
First, the legal questions are moving upstream, from output to acquisition. The Suno amendment concerns how recordings were obtained and whether an access control was circumvented, not only whether protected works were used. The two Round Hill filings concern the provenance of training inputs. The OpenAI sanctions dispute concerns preservation and whether training and output evidence can be reconstructed at all. The Apple derivative action asks whether directors oversaw those same acquisition practices. A year of litigation about what generative systems produce is being joined by litigation about how they were built.
Second, AI assistants are being tested under conventional privacy law rather than AI-specific law. The Otter ruling applies federal and state wiretap statutes and an Illinois biometric statute to a meeting notetaker. The Grok class action, which will now proceed in California, concerns whether conversations with a chatbot were transmitted to third parties through tracking technology. Neither case required a new statute. The question in both is a familiar one asked of an unfamiliar participant: when an AI service is present in a conversation, is it the user's tool or a third party in its own right?
Third, synthetic identity is being approached from several directions at once, and the answers are not converging.Brazil's electoral court treated the synthetic and undisclosed character of the material as central to interim relief. A São Paulo civil judge, on a different record and under different law, declined emergency removal of an AI-generated political video on the basis that it appeared to be satire and political criticism. The Babylon Bee's federal complaint contends that requiring disclosure on AI-manipulated political material captures protected parody. Indian courts continue to grant interim personality-rights relief over synthetic depictions. Whether the law regulates the fact that something is synthetic, or only deception and harm, remains open.
Fourth, remedies directed at models and systems remain a live question rather than a settled one. The multistate trial against Meta has now reached evidence, with the states seeking, among other relief, deletion of algorithms or AI models said to have been developed using children's data. No such remedy has been granted. It follows the New Mexico remedial judgment reported in Issue 002, in which AI-related operational conduct formed part of court-ordered relief. Structural relief of this kind is being sought in contested litigation rather than agreed in settlement, which is what makes the trial worth following.
Where primary documentation could not be obtained during the reporting period, that limitation is stated in the entry concerned rather than the matter being omitted. Every entry carries both a verification status and a statement of the source actually seen.
LEAD DEVELOPMENTS
Vietnam's list of high-risk AI systems enters into force
Jurisdiction: Vietnam — national Authority: Prime Minister / Government of Vietnam Instrument: Decision No. 33/2026/QĐ-TTg — List of High-Risk Artificial Intelligence Systems Area: AI regulation · Healthcare · Education · Financial services · Biometrics · Transport · Judicial and legal proceedings Stage: Regulatory instrument in force Date:15 August 2026
ALR-0070 · Verification: Verified | Source basis:Primary — Government of Vietnam legal-document portal and official gazette record reviewed
What happened. Decision No. 33/2026/QĐ-TTg was issued on 30 June 2026 and took effect on 15 August. It promulgates the national list of artificial-intelligence systems classified as high risk. The listed areas include education, ethnicity and religion, healthcare, banking, legal and judicial proceedings and transport. Examples identified in the government's own explanatory material include AI used to assess or rank students automatically, to decide credit, to support robotic surgery, to perform certain large-scale biometric functions in proceedings, and to operate higher levels of autonomous transport or critical transport infrastructure.
The instrument requires human supervision, control and the capacity to intervene in the operation of listed systems, and states that the use of AI does not remove the legal authority or responsibility of the competent person or body.
The underlying question. Classification determines obligation. The practical question a high-risk list answers is which uses of AI attract heightened duties, and the more interesting one is what those duties consist of: Vietnam's answer places human supervision and non-delegable institutional responsibility at the centre rather than, for example, ex ante approval. The instrument is a classification decision made under Vietnam's wider AI legal framework; it is not itself that framework, and the 15 August event should not be described as the enactment of Vietnam's AI Law.
What happens next. Transitional periods apply to systems already operating before commencement. On the government's explanation, listed systems in healthcare, education and finance generally have until 1 September 2027; other pre-existing listed systems generally have until 1 March 2027, subject to the decision's detailed provisions. Implementation guidance and any sector-specific enforcement remain to be identified.
Legal point. An operative national high-risk classification instrument now expressly provides that the use of an AI system does not displace the legal responsibility of the person or body otherwise competent to decide.
Core privacy claims survive dismissal in the Otter.AI meeting-assistant litigation
Jurisdiction: United States — Northern District of California (federal) Court: US District Court for the Northern District of California (Judge Eumi K. Lee) Case:In re Otter.AI Privacy Litigation, No. 5:25-cv-06911-EKL Area:Privacy · Wiretap and interception law · Biometric data · Unfair competition · Model training Stage: Motion to dismiss denied in material part; consolidated proposed class action continuing Date: 13 August 2026
ALR-0071 · Verification: Verified | Source basis:Primary as to the existence and date of the order from the federal filing record; full order text not extracted. Secondary — Bloomberg Law for the disposition of individual counts
What happened. The consolidated proceedings arise from allegations that Otter.ai's AI meeting assistant joins virtual meetings, records and transcribes conversations involving participants who may not themselves be Otter users, and then retains or processes those communications, including in connection with machine-learning systems. Several earlier suits were consolidated in the Northern District of California.
On 13 August the court denied Otter's motion to dismiss in material part. Reporting of the order records that claims under the federal Electronic Communications Privacy Act, California wiretap provisions, the Illinois Biometric Information Privacy Act, the California Unfair Competition Law and unjust enrichment were allowed to continue.
The underlying question. Whether an AI meeting assistant that records and transcribes a conversation is properly characterised as the instrument of the participant who invited it, or as a separate party to the communication. Interception statutes generally turn on that distinction, and consent obtained from one participant does not necessarily answer it for the others. The biometric claims raise a second question: whether processing voice data for transcription and model-related purposes involves the kind of identifier those statutes were written to protect.
What this is not. A pleading-stage ruling. The court has decided that the claims as pleaded may proceed, not that Otter intercepted communications unlawfully, collected biometric information unlawfully, or is liable to anyone.
What happens next. Plaintiffs may amend any claims dismissed with leave to amend; responsive pleading and class-action proceedings follow. Precise deadlines require the full order.
Legal point. Interception and biometric statutes drafted long before generative AI are being applied to AI assistants that participate in conversations rather than merely store them.
Record companies permitted to add DMCA anti-circumvention claim against Suno
Jurisdiction: United States — District of Massachusetts (federal) Court: US District Court for the District of Massachusetts (Chief Judge F. Dennis Saylor IV) Case:UMG Recordings, Inc. et al. v Suno, Inc. et al., No. 1:24-cv-11611 Area: Copyright · DMCA §1201 · Model training · Data acquisition Stage: Leave to amend granted; claim to be pleaded Date: 18 August 2026
ALR-0072 · Verification: Verified | Source basis:Primary — federal docket confirms the canonical action, case number and judge. Secondary — Bloomberg Law and contemporaneous docket-based analysis for the terms of the 18 August order; signed order not extracted
What happened. The copyright action against the generative-music service Suno has been pending since June 2024. On 18 August the court permitted the remaining record-company plaintiffs to amend to add a claim under 17 U.S.C. §1201(a), the anti-circumvention provision of the Digital Millennium Copyright Act. The proposed claim alleges that Suno obtained sound recordings by "stream ripping" from YouTube, circumventing a technological measure controlling access to those recordings in order to obtain training material.
The court also declined, at this stage of the litigation, to allow more than 61,000 additional works to be added to the existing action.
The underlying question. Two questions, and they are separable. The narrow one is whether YouTube's rolling cipher is a measure that controls access to a work within §1201(a), or only a measure that controls copying — a distinction with a long history in circumvention litigation. The broader one is whether liability in AI-training cases can attach to the method by which training material was acquired, independently of whether the eventual use of that material was infringing. A fair-use defence addresses the second question; it does not obviously address the first.
What this is not. The court did not find that Suno stream-ripped recordings or violated the DMCA. It concluded that the theory could not be rejected on the pleadings and allowed it to be litigated.
What happens next. The plaintiffs file the permitted amended pleading; the deadline and Suno's responsive timetable require docket extraction. This matter is distinct from Round Hill Music v Suno (ALR-0076), filed the previous day in California.
Legal point. AI-training litigation is extending from the use of protected works to the means by which they were acquired, where circumvention is alleged.
Multistate youth-safety trial against Meta reaches opening statements
Jurisdiction: United States — Northern District of California (federal) Court: US District Court for the Northern District of California, Oakland (Chief Judge Yvonne Gonzalez Rogers) Case: Multistate Attorneys General litigation against Meta Platforms Area: Consumer protection · Child safety · Recommendation systems · Children's data · Structural remedies Stage: Merits trial underway Date: 18 August 2026
ALR-0039 (update) · Verification: Partial | Source basis:Secondary — contemporaneous courtroom reporting by Reuters and AP; no trial minute, transcript or trial-track docket number retrieved
What happened. Issue 002 recorded the start of jury selection on 12 August. Opening statements took place on 18 August in the federal trial arising from claims brought by 29 states, with California, Colorado, Kentucky and New Jersey leading the present phase. The states allege that Meta designed platform features to maximise engagement among young users, misrepresented aspects of platform safety and unlawfully handled children's data. Meta denies the allegations. Evidence began immediately, with former Meta engineer Arturo Béjar testifying on the opening day about internal child-safety concerns; that is testimony offered by the plaintiffs, not a judicial finding.
Where AI enters. Principally through the remedies sought. As recorded in Issue 002, the states seek, among other relief, deletion of algorithms or AI models allegedly developed using children's data, together with changes to recommendation and ranking systems. No such relief has been granted, and the exact operative wording of the remedy sought in the current trial papers has not been verified.
The underlying question. Whether a court, on a contested record, will order the deletion of models and algorithms as a remedy — algorithmic disgorgement — rather than accepting it as a negotiated term. The remedy has appeared in earlier federal consumer-protection settlements; it is rarely tested at trial.
What happens next. The trial is expected to run for several weeks. Mark Zuckerberg and Adam Mosseri are expected to testify; no fixed dates have been verified. The proceedings are legally distinct from the New Mexico matter at ALR-0025.
Brazil's electoral court orders removal of AI-generated video in a new representation
Jurisdiction: Brazil — federal electoral jurisdiction Court: Tribunal Superior Eleitoral (Minister André Mendonça) Case: Representation No. 0601617-29.2026.6.00.0000 Area: Electoral law · Synthetic media · Platform disclosure obligations Stage: Interim relief granted; representation pending Date: 20 August 2026
ALR-0073 · Verification: Verified | Source basis:Primary in indexed form — TSE PJe representation record confirms the case number, court and date; the rendered decision is behind the court's access control. Secondary — CNN Brasil, UOL and other contemporaneous Brazilian reporting for the terms of the order
What happened. The representation concerns an AI-generated video published by the Brasil Sátira do Poder account, depicting presidential candidate Flávio Bolsonaro and the banker Daniel Vorcaro in fabricated, photorealistic scenes. On 20 August Minister André Mendonça ordered the responsible account and Meta to remove the material within 24 hours and prohibited republication. Meta was ordered to provide account information, and the fundraising platform Vakinha was directed to provide specified information concerning an associated campaign.
Reporting of the interim reasoning indicates that the synthetic character of the material was central: describing the piece as satire was not treated as removing it from the electoral rules governing realistic synthetic content where the use of AI is not prominently disclosed.
The underlying question. Where the line falls between satire and regulated synthetic electoral content when the material is photorealistic. Brazil's electoral rules require the use of AI to be disclosed in political material; the question in this proceeding is whether an intention to satirise displaces that requirement, or whether realism engages it regardless of intent. That is closely related to, but procedurally separate from, the question pending in the Flávio Bolsonaro proceeding at ALR-0033.
What this is not. Interim relief. It does not finally determine whether the publication constitutes unlawful negative early electoral propaganda, nor resolve the merits of the representation.
What happens next. Defence submissions, further consideration of the alleged early negative propaganda, and plenary review of the interim decision. No merits date has been retrieved.
Legal point. An electoral court has treated the photorealism and non-disclosure of synthetic content, rather than the publisher's satirical intent, as decisive for interim relief.
Babylon Bee challenges New Mexico's AI political-media law on First Amendment grounds
Jurisdiction: United States — District of New Mexico (federal) Court: US District Court for the District of New Mexico Case:The Babylon Bee LLC v Castillo, No. 26-cv-2628 Area: Constitutional law · Free expression · Synthetic political media · Disclosure requirements Stage: Complaint filed Date: 11 August 2026
ALR-0074 · Verification: Verified | Source basis:Primary in part — complaint and docket identified through docket-linked authoritative reporting; full complaint not separately extracted
What happened. The satirical publisher The Babylon Bee filed a federal action challenging New Mexico's statutory requirements for disclaimers on certain "materially deceptive media", including political material generated or manipulated using artificial intelligence. The Bee alleges that applying those requirements to political satire and parody infringes the First Amendment, and seeks declaratory and injunctive relief preventing enforcement against its intended content. The New Mexico Ethics Commission disputes the characterisation and says the provision has not been enforced against parody or satire.
The underlying question. Whether a state may compel disclosure on AI-manipulated political communications without capturing expression that is protected because it is understood by its audience not to be true. Disclosure regimes are ordinarily easier to defend than prohibitions, and this action tests that proposition at its most awkward point: satire depends on the audience recognising artifice, which is close to the interest a disclosure requirement asserts.
What this is not. A ruling. The complaint is the event. No court has determined that the New Mexico statute is unconstitutional, and no finding has been made that the Commission has enforced it unlawfully.
What happens next. Defendants' response and any preliminary-injunction proceedings. The assigned judge and response schedule have not been established. The matter should be read alongside the separate xAI challenge to Minnesota's synthetic-intimate-imagery statute (ALR-0009): both concern how far a state may regulate AI-generated content without reaching protected expression, but they involve different statutes, different content and different constitutional arguments.
Legal point. The first significant constitutional test of AI political-disclosure rules in this record arises not from deceptive content but from content intended to be recognised as false.
COURTS & LITIGATION
Two new music-publisher actions were filed on the same day in the same court, one against a foundation-model developer and one against a generative-music service. They are reported separately because they concern different defendants, different works and different systems, and will follow separate procedural paths.
United States — Round Hill Music sues Anthropic over lyrics used in training
US District Court, Northern District of California · Round Hill Music LP et al. v Anthropic PBC, No. 5:26-cv-08505-NC · Event 17 August 2026 · Complaint filed; initial case-management scheduling order entered
ALR-0075 · Verification: Verified | Source basis:Primary — the Northern District of California filing index confirms the case, number and same-day docket activity; complaint not separately extracted. Secondary — Reuters
Round Hill Music alleges unauthorised copying of lyrics from at least 500 compositions for use in training Anthropic's generative-AI systems, including Claude. These are allegations; no infringement has been established and no defence has yet been filed.
The court entered an initial case-management scheduling order with ADR deadlines on the same day the complaint was filed. That is procedural and says nothing about the merits.
The action is factually related to earlier music-publisher litigation against Anthropic but is a separate proceeding. Next steps are service, responsive pleading and case management; exact deadlines require docket extraction.
United States — Round Hill Music sues Suno over compositions used in training
US District Court, Northern District of California · Round Hill Music LP et al. v Suno Inc., No. 5:26-cv-08507 · Event 17 August 2026 · Complaint filed
ALR-0076 · Verification: Partial | Source basis:Secondary — Reuters identifies the court, filing and case number; complaint not retrieved
Round Hill separately alleges that copyrighted compositions were copied without authorisation to train Suno's generative music system, with a catalogue initially said to include at least 500 compositions and capable of expansion. No liability has been determined.
The proceeding is distinct from UMG Recordings v Suno (ALR-0072) in Massachusetts, which concerns different plaintiffs and has now acquired a separate circumvention theory. Taken together, the two actions place Suno's training practices before two federal courts on overlapping but not identical legal footings.
United States — discovery sanctions contested in the OpenAI copyright MDL
US District Court, Southern District of New York · In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-03143 · Events 13 and 14 August 2026 · Sanctions application opposed; determination pending
ALR-0080 · Verification: Verified | Source basis:Primary in part — the filings are identified and linked through docket-based reporting by Bloomberg Law, corroborated by Law360; full briefs not extracted
The consolidated litigation brought by The New York Times and other publishers against OpenAI has produced a substantial discovery dispute. The publishers contend that OpenAI misrepresented its ability to search training material and model-output logs, failed adequately to preserve some user conversations, and obstructed attempts to establish whether ChatGPT reproduced their works. They seek sanctions that would, among other things, restrict OpenAI's use of a sample of roughly 20 million ChatGPT conversations and secure evidential consequences relating to alleged regurgitation.
OpenAI filed its opposition on 14 August, denying discovery misconduct and arguing that it complied with preservation obligations, provided training data and large volumes of output material, and that excluding the disputed sample would materially impair its defence. Microsoft, a co-defendant against whom sanctions are not sought, filed a separate response on 13 August arguing that the relief sought against OpenAI could prejudice its own ability to defend the consolidated claims, particularly if OpenAI were prevented from relying on the output sample.
The substance here is evidential rather than doctrinal, and it matters for that reason. Whether a copyright claim against a generative system can be proved at all depends on whether training datasets and output logs can be searched, preserved and produced in a form a court can work with. Both sides' accounts remain contested and no sanctions have been imposed. Next step is the court's determination; no ruling date has been located.
Legal point. The practical viability of AI-training copyright claims is being litigated through preservation and searchability of model inputs and outputs, in advance of any merits determination.
United States — Zuckerberg deposition permitted in coordinated Meta AI copyright suits
US District Court, Northern District of California (Judge Vince Chhabria) · Entrepreneur Media, LLC v Meta Platforms, Inc., No. 3:25-cv-09579, and coordinated actions · Event 11 August 2026 · Discovery order (backfilled)
ALR-0081 · Verification: Partial | Source basis:Docket confirmed as to the coordinated proceedings and the deposition deadline; complete 11 August order across the coordinated actions not extracted. Secondary — Bloomberg Law
The court rejected Meta's attempt to prevent Mark Zuckerberg being deposed in a group of copyright suits concerning the acquisition and use of copyrighted material in developing Meta's AI systems. The newer actions, brought by publishers and authors including Entrepreneur Media, have been related to Kadrey v Meta for discovery purposes. Plaintiffs contend that Zuckerberg has firsthand knowledge of Meta's AI-training initiatives, pointing to alleged knowledge of a programme described in the litigation as involving the acquisition and scanning of physical books for training. Those remain litigation allegations.
The ruling is a discovery ruling. It determines nothing about infringement, fair use or the lawfulness of Meta's training practices. The deposition is to occur no later than 26 August 2026.
United States — Apple shareholder derivative action alleges failure to oversee AI legal risk
US District Court, Northern District of California · Rosen v Cook et al., No. 5:26-cv-08463 · Event 14 August 2026 · Derivative complaint filed (backfilled)
ALR-0082 · Verification: Partial | Source basis:Docket identity and filing date corroborated through specialist docket reporting (Law360); complete complaint not extracted
An Apple shareholder filed a derivative action against Apple officers and directors alleging that they exposed the company to legal and reputational risk through its AI-training practices, including alleged use of copyrighted or otherwise unauthorised source material. These are shareholder allegations. No court has found that Apple infringed copyright or that its directors breached any fiduciary duty.
The matter is recorded because it introduces a category of AI litigation not previously captured in this record: not whether an AI system caused harm, but whether a board discharged its oversight duty in relation to how AI systems were developed. Derivative claims of this kind are difficult to sustain and frequently fail at the pleading stage. Their significance for boards lies less in their prospects than in what they make discoverable.
Next steps are responsive pleading and derivative-litigation scheduling; no fixed date is verified.
Legal point. AI-training practices are beginning to be pleaded as a matter of director oversight, distinct from the underlying copyright claims themselves.
United States — Grok privacy class action remains in California
US District Court, Northern District of California (Judge Beth Labson Freeman) · Skaggs v X.AI, LLC, No. 5:26-cv-04550 · Event 13 August 2026 · Transfer refused
ALR-0077 · Verification: Partial | Source basis:Docket identity, case number, judge and underlying filings independently established; the 13 August order not directly retrieved. Secondary — Law360 for the terms of the ruling
The proposed class action, filed in May 2026, alleges that conversations between users and xAI's Grok chatbot, including potentially sensitive information entered into the service, were transmitted through tracking technology to third parties including Google, Meta and TikTok. The claims are reported to include federal and California privacy causes of action. These are allegations; no privacy violation has been established.
On 13 August the court refused xAI's application to transfer the action to Texas. The immediate issue was forum, not whether Grok unlawfully disclosed users' communications, and the ruling determines nothing about the merits or about certification of any class.
The action proceeds in the Northern District of California. No new fixed date has been located.
United States — Court of Federal Claims rejects Scale AI's bid protest
US Court of Federal Claims (Judge Zachary N. Somers) · Scale AI, Inc. v United States, No. 1:26-cv-00155 · Event 18 August 2026 · Post-award bid-protest decision
ALR-0084 · Verification: Partial | Source basis:Primary as to the docket, parties, judge and procedural history, including merits motions on the administrative record and oral argument on 20 May. Secondary — specialist Court of Federal Claims reporting for the 18 August outcome; the merits opinion has not been obtained
Scale AI brought a post-award Tucker Act protest challenging a National Geospatial-Intelligence Agency procurement awarded to Enabled Intelligence, which intervened. The procurement concerns data-labelling and related support for government AI and machine-learning activity. Reporting states that the court rejected Scale AI's challenge on 18 August.
Detailed reasoning is withheld pending review of the opinion. The Report can record that the protest was rejected; it cannot yet state which grounds were considered or how they were resolved, whether separate judgment was entered, or what redactions apply. Any wider relationship to other defence AI programmes should not be assumed from reporting alone.
The decision concerns the conduct of a federal procurement. It is not a determination about the quality, safety or technical merits of either party's AI capability. The matter is recorded because procurement of the data infrastructure on which government AI depends rarely produces a public judicial record, and this record is now capable of being followed. Any appeal to the Federal Circuit remains to be identified; the transcript of the 20 May oral argument is scheduled to come off restriction on 21 September 2026.
United States — Minnesota opposes xAI's injunction application in the nudification challenge
US District Court, District of Minnesota · X.AI LLC v Keith Ellison, No. 26-cv-03425-DWF-DTS · Events 14 and 19 August 2026 · State opposition filed; preliminary-injunction hearing held 19 August; ruling awaited
ALR-0009 (update) · Verification: Partial | Source basis:Primary as to the procedural timetable — the court's 31 July order converted the emergency application into a preliminary-injunction motion and fixed the 19 August hearing. The 14 August opposition brief was not extracted, and the hearing itself is established from contemporaneous Minnesota reporting; no transcript or minute retrieved. Secondary — Reuters and Minnesota court reporting
Issue 002 recorded amicus support for xAI's challenge to Minnesota's statute restricting AI tools used to generate or alter intimate imagery. The Attorney General filed the State's opposition to the preliminary-injunction application on 14 August, arguing that the statute is directed at harmful synthetic intimate imagery. xAI maintains that the law is overbroad and reaches protected expression.
Oral argument on the preliminary-injunction application took place on 19 August. Contemporaneous reporting confirms that the parties argued the constitutional challenge before Judge Donovan Frank. No ruling was issued at the hearing, and no subsequent order has been located as at 21 August. Local reporting following the hearing stated that a decision was not expected until at least the following week.
Neither the State's opposition nor the hearing resolves the constitutional question. Minnesota's law remains in force unless and until the court orders otherwise.
India — Delhi High Court grants interim personality-rights relief to Khushi Kapoor
Delhi High Court (Justice Jyoti Singh) · Khushi Kapoor v Ashok Kumar/John Doe & Ors., CS(COMM) 869/2026 · Event 12 August 2026 · Ex parte ad-interim injunction (backfilled)
ALR-0078 · Verification: Verified | Source basis:Primary — digitally signed Delhi High Court order downloaded from the court's server
The proceedings concern alleged unauthorised exploitation of the claimant's identity and personality attributes. The material before the court included alleged AI-generated obscene imagery, deepfakes, digitally manipulated content and AI chatbot characters. The court granted interim restraints and takedown directions while the action proceeds, and directed summons to the defendants.
The order is notable within this record for the breadth of technology it expressly engages: generative AI, deepfakes, AI chatbots, face morphing, voice cloning and synthetic media. Services including SeaArt AI and JanitorAI are among those identified in the order and reporting materials, alongside platform and search intermediaries; no single foundation model is established as the source of all the complained-of content.
This is interim relief. It is not a determination that any defendant infringed the claimant's rights or created any particular item of synthetic content. The matter is legally separate from the Tabu and Janhvi Kapoor proceedings recorded in Issue 002, though the Delhi High Court is plainly now working through a recurring set of questions about the scope of such orders. Next date: 17 September 2026 before the Joint Registrar.
India — Chandigarh court orders 72-hour takedown of alleged AI-manipulated political video
Civil court, Chandigarh (Civil Judge Kaushal Kumar Yadav, as reported) · Sukhpal Singh Khaira v Charda Punjab / John Doe and Meta · Event 12 August 2026 · Ad-interim injunction
ALR-0079 · Verification: Partial | Source basis:Secondary — detailed specialist legal reporting and Times of India court reporting; no signed order, and the official case number has not been obtained
The Punjab MLA Sukhpal Singh Khaira sued over a video distributed through Facebook and Instagram accounts which he alleges used artificial intelligence to place his recognisable face, beard and turban onto an artificially generated body. The account operators are presently treated as unidentified John Doe defendants, with Meta joined in relation to removal and disclosure.
Reporting states that the court found a prima facie basis for interim protection and directed Meta to remove or disable access to the identified URLs within 72 hours, on the basis of Order XXXIX Rules 1 and 2 read with section 151 of the Code of Civil Procedure. Fresh notices were issued to the unidentified operators. Khaira has sought subscriber-identification information; the material reviewed does not establish that the court finally granted that relief at this stage.
The ruling is ad interim and rests on a prima facie assessment. It is not a determination of defamation, of authorship of the synthetic video, or of Meta's liability. Next date: 14 September 2026 for compliance, written statement and response to the interim application, as reported.
Brazil — two São Paulo civil actions over AI-generated political videos produce different interim results
41st Civil Court of São Paulo (as reported) · Fábio Luís Lula da Silva v Flávio Bolsonaro and Fábio Luís Lula da Silva v Romeu Zema · Event 18 August 2026 · Civil complaints filed; emergency relief refused in one
ALR-0085 and ALR-0086 · Verification: Partial | Source basis:Secondary — Folha de S.Paulo/Folhapress and corroborating São Paulo reporting; no case numbers, complaints or signed decisions retrieved
Two civil actions were filed on the same day by the same claimant over separate AI-generated political videos. They are recorded as separate matters.
In the first (ALR-0085), the claimant challenges a video published through Flávio Bolsonaro's social-media accounts, generated using AI, depicting the defendant piloting an aircraft on a fictional mission while making allegations connected to the INSS fraud investigations. The claimant seeks removal from X and Instagram within 24 hours, restraint against republication, R$10,000 in moral damages and a finding that the use of AI to communicate what he says are false imputations constitutes an abuse of rights. The material retrieved does not establish a decision on the urgent application.
In the second (ALR-0086), concerning an AI-generated video published by Romeu Zema, the judge reportedly declined to order immediate removal, on the basis that the material appeared on the initial record to fall within satire and political criticism and that manifest abuse or obviously unlawful content had not been established to the standard required for exceptional immediate takedown.
The contrast with the TSE order at ALR-0073 is worth noting carefully, because it is not a conflict. Different courts applied different bodies of law: electoral rules that expressly regulate undisclosed synthetic content in one case, and general civil law governing personality rights and expression in the other. Refusal of emergency removal is interlocutory and does not establish that the video is lawful. All allegations remain the claimant's case; no finding has been made against either defendant.
Philippines — AI-generated demonstrative material used in the Duterte impeachment trial
Senate of the Philippines sitting as an impeachment court · Impeachment proceedings concerning Vice President Sara Duterte · Event 17 August 2026 · Use of AI-generated material during witness examination; challenge raised
ALR-0083 · Verification: Partial | Source basis:Traceable to the official Senate livestream of the 17 August hearing; no written Senate transcript or evidentiary ruling retrieved. Secondary — Philippine Daily Inquirer and BusinessMirror
During examination of a former Office of the Vice President special disbursing officer, the House prosecution presented AI-generated images intended to visualise what ₱125 million in cash might look like when bundled and placed in bags. The prosecutor disclosed that AI had been used to create the images. The defence challenged their accuracy and provenance, and the witness corrected aspects of what was shown, including how the bags were depicted. The prosecution's position, as subsequently reported, was that the images were a visual representation rather than independent proof of the underlying transaction.
The underlying question. What status generated imagery has when used to illustrate testimony in a tribunal. Demonstrative aids have long been permitted precisely because they are not evidence of the facts depicted; the difficulty with a photorealistic generated image is that the distinction is harder for an audience to hold, and the material carries an implicit claim to accuracy that a diagram does not.
What this is not. There was no general ruling on the admissibility or evidential status of AI-generated material, and the images must not be described as proof that money was packaged or transported as depicted. The proceedings are constitutional rather than criminal, and no finding has been made.
The trial continues. No AI-specific hearing date has been identified.
Legal point. AI-generated demonstrative material is entering formal proceedings in advance of any settled rule about how tribunals should treat it.
REGULATORS, ENFORCEMENT & OVERSIGHT
England and Wales — Charity Commission places AI within the sector risk environment
Charity Commission for England and Wales · Charity Sector Risk Assessment 2026 · Event 18 August 2026 · Regulatory risk assessment published
ALR-0087 · Verification: Verified | Source basis:Primary — Charity Sector Risk Assessment 2026, published 18 August, reviewed
The Commission published its 2026 sector risk assessment, which addresses artificial intelligence in two distinct contexts. First, it identifies the use of AI in charity-registration applications as a means by which bad actors may seek to exploit the registration system. Second, in its treatment of emerging technology and cyber risk, it records that AI can benefit charities but may increase risk to beneficiaries and employees if misused.
The operative point for trustees is that the Commission states they remain responsible for ensuring appropriate safeguards, oversight and risk management where AI is used.
The assessment creates no new statutory duty. It applies existing trustee responsibilities to AI-related risk and records the regulator's view of where that risk sits. It is recorded here because it is an example of a general regulator, without any AI-specific mandate, formally locating AI within the risk environment of the sector it supervises — the point at which AI use becomes a governance question for the regulated body rather than an operational choice. No enforcement action or finding against any charity arises from the publication. No further date is fixed.
Legal point. Existing trustee and officer duties are being applied to AI use by general regulators without new AI-specific powers.
United States — NAIC AI Risk Evaluation Supplement pilot confirmed by official summary
NAIC Big Data and Artificial Intelligence (H) Working Group · Event 13 August 2026 · Working-group session; multistate pilot continuing. Institutional status: state-insurance regulatory coordinating body, not a regulator in its own right
ALR-0051 (update) · Verification: Verified | Source basis:Primary — NAIC official Meeting Summary Report dated 14 August 2026
Issue 002 reported this matter on the basis of prepared agenda and pilot materials only, and expressly qualified that no post-meeting record had been located. That gap has now closed.
The official summary confirms that the Working Group met on 13 August and received an update on the AI Risk Evaluation Supplement pilot. It records that the pilot began in March, involves twelve participating states, continues through September, and is being used in market-conduct examinations, financial examinations, financial analyses and broader regulatory inquiries. The Working Group also heard a presentation on AI governance in insurance covering predictive, generative and agentic AI, validation, human oversight, monitoring, documentation, authorisation limits and rollback controls.
The verification status of the 13 August event is accordingly upgraded from Partial to Verified. The qualification stated in Issue 002 nevertheless stands in one respect: no source states that the Supplement itself was adopted at that meeting. A further iteration is anticipated at the end of August, with the current pilot period continuing through 30 September 2026.
Brazil — ANPD generative-AI and data-protection session takes place
Agência Nacional de Proteção de Dados · 3º Encontro ANPD de Encarregados · Event 19 August 2026 · Scheduled event completed
ALR-0052 (update) · Verification: Verified | Source basis:Primary — ANPD official event page and programme; Ministry of Justice and Public Security post-event account published 19 August
Issue 002 recorded the publication of the programme for the national data-protection authority's third meeting of data-protection officers, including a session on personal data and the use of generative AI in public and private organisations. The meeting took place in Brasília on 19 August, and the government's post-event account confirms that generative AI was substantively discussed, alongside security incidents and the interaction between the LGPD and Brazil's digital framework for children and adolescents.
No resulting resolution, guidance or other regulatory instrument has been identified. The event was regulator-led professional engagement. It did not create binding guidance, impose any sanction or amend the LGPD, and it should not be reported as having done so. No further step has been identified.
GOVERNMENT & INSTITUTIONAL DEVELOPMENTS
Türkiye — AI Action Plan 2026–2030 moves into implementation by Presidential Circular
Presidency of the Republic of Türkiye / Ministry of Industry and Technology · Presidential Circular 2026/9; Official Gazette No. 33344 · Event 18 August 2026 · Circular published; government implementation required
ALR-0088 · Verification: Verified | Source basis:Primary in substance — official legislation notices confirm Circular 2026/9, Official Gazette No. 33344 and publication on 18 August; the Gazette's rendered text was not directly returned. Corroborated by Anadolu Agency
The Circular, dated 17 August and published on 18 August, concerns the Türkiye Artificial Intelligence Action Plan 2026–2030 and requires public institutions and organisations to carry out the responsibilities allocated to them under the Plan and to provide the support necessary during implementation. The Plan is coordinated by the Ministry of Industry and Technology and is stated to rest on principles including human-centredness, trustworthiness, ethical responsibility, digital sovereignty and sustainable development. It addresses AI literacy and skills, data and computing infrastructure, public- and private-sector adoption, and the development of a competitive and trustworthy AI ecosystem.
The date matters. The Plan itself was announced by the Ministry on 13 June 2026; the 18 August event is the formal instrument through which implementation obligations are placed on public bodies. It is not the first announcement of the Plan, and it is neither an enforcement action nor a finding of non-compliance against anyone.
Implementation runs to 2030. Individual milestones require extraction of the Plan itself.
New South Wales — Office of AI to be established within the Cabinet Office
NSW Government / The Cabinet Office · Event 17 August 2026 · Institution announced; implementation to follow
ALR-0089 · Verification: Verified | Source basis:Primary — NSW Government ministerial release dated 17 August 2026
The NSW Government announced that it will establish an Office of AI within The Cabinet Office, intended to strengthen the state's response to AI, coordinate with Digital NSW on government investment and adoption, and work with public-sector unions on responsible AI use. The announcement accompanied the state's Data Centre Policy Framework and related infrastructure measures.
The characterisation requires care. This is a central government coordination and oversight unit situated within the Cabinet Office. It is not an independent statutory AI regulator, and the announcement does not itself create one; that would require a later instrument. Commencement date, staffing and detailed mandate have not been established.
The Office joins a growing set of state and national AI coordination bodies recorded in this Report, including the AI Office of Ireland (ALR-0054). It is separate from the NSW school-assessment review at ALR-0055.
Ireland — consultation opens on the EU Cloud and AI Development Act
Department of Enterprise, Tourism and Employment · Event 14 August 2026 · Public consultation opened (backfilled)
ALR-0090 · Verification: Verified | Source basis:Primary — Department consultation page reviewed
The Department opened a public consultation on the European Commission's proposed Cloud and AI Development Act, described as a proposed framework intended to strengthen Europe's cloud-computing and AI ecosystem.
The consultation is an Irish evidence-gathering exercise intended to inform the Government's position in the European legislative process. It is not proposed Irish domestic AI legislation, and it is not adoption or enactment of the European proposal, which remains at Commission stage.
Submissions close on 11 September 2026.
LEGISLATION & RULEMAKING
Colorado — Attorney General files proposed automated decision-making and chatbot rules
Colorado Department of Law / Office of the Attorney General · Proposed Automated Decision-Making Technology & Conversational Artificial Intelligence Service Rules · Event 11 August 2026 · Proposed rules filed; formal rulemaking open (backfilled)
ALR-0091 · Verification: Verified | Source basis:Primary — Colorado Attorney General official rulemaking record, including the rulemaking notice and statement of basis, authority and purpose
The Department of Law filed proposed rules with the Colorado Secretary of State on 11 August. They are intended to clarify and implement two statutes: SB 26-189, governing automated decision-making technology used to materially influence consequential decisions, and HB 26-1263, the Chatbot Safety Act.
As described in the Attorney General's materials, the chatbot provisions address age estimation, disclosure that a user is interacting with AI rather than a human, protections for minors against specified sexual content and simulated emotional dependence, suicide and self-harm protocols, and reporting to the Attorney General.
What this is not. These are proposed rules. They are not final regulations, and nothing takes effect on the strength of the filing. Both underlying statutory schemes are stated to become operative on 1 January 2027. The rulemaking is separate from the Colorado psychotherapy AI statute recorded at ALR-0062, which concerns different legislation and different regulated activity.
What happens next. Written comments are open through 26 October 2026 at 23:59 MST, subject to extension if the formal hearing continues beyond that date. The Attorney General asks that comments intended to inform revisions presented at the rulemaking hearing be submitted by 5 October 2026.
Legal point. Implementation detail — age estimation, disclosure, self-harm protocols — is being settled through subordinate rulemaking rather than in the statutes themselves.
Thailand — consultation closes on the revised draft AI Act
Electronic Transactions Development Agency (ETDA) · Event 14 August 2026 · Public consultation closed
ALR-0092 · Verification: Partial | Source basis:Secondary — multiple specialist legal sources independently record the 9 July release and 14 August deadline; the ETDA consultation notice itself was not retrievable through the current search index
ETDA released a revised draft Artificial Intelligence Act for public consultation in July 2026, with comments open until 14 August. The draft proposes a dedicated risk-based framework, with differentiated obligations according to the risk presented by a system and mechanisms including regulatory sandboxes.
The event is closure of a consultation on draft legislation. Thailand has not enacted an AI Act, and the draft's contents remain subject to change. The Report has not obtained the primary deadline record; the entry is published with that gap stated rather than omitted, since the underlying legislative exercise is central to Thailand's approach to AI.
Next step is government consideration of responses and any revised legislative proposal. No Cabinet or legislative date has been verified.
Italy — parliamentary-opinion deadline passes on the EU AI Act implementing decree
Italian Parliament / Senate committees; Government delegated-legislation process · Atto del Governo No. 421 · Event 16 August 2026 · Opinion deadline expired
ALR-0093 · Verification: Verified as to the deadline only | Source basis:Primary — Italian Senate parliamentary record
Atto del Governo No. 421 concerns a draft legislative decree adapting Italian law to Regulation (EU) 2024/1689, the EU AI Act. The parliamentary record establishes 16 August 2026 as the deadline for the relevant parliamentary opinion, and that deadline has now passed.
The verified event is the expiry of the deadline and nothing more. It has not been established that a committee opinion was adopted, what any such opinion said, or that the Government has since adopted or published the final decree. The entry is deliberately narrow: the passage of a scrutiny deadline is a procedural fact, and it would be wrong to convert it into a parliamentary outcome.
Next step is confirmation of any committee opinion and of subsequent Government adoption. No date has been confirmed.
DEVELOPMENTS TO WATCH
Twenty-four live matters awaiting their next trigger, with the position as at 21 August 2026.
Imminent
ALR-0039 Multistate youth-safety trial v Meta (N.D. Cal.) — Opening statements 18 August; evidence underway; AI-model deletion sought as remedy. Next: Trial continues; Zuckerberg and Mosseri expected to testify, dates unverified.
ALR-0081 Coordinated Meta AI copyright suits (N.D. Cal.) — Deposition permitted 11 August. Next: Zuckerberg deposition to occur by 26 August 2026.
ALR-0009 X.AI LLC v Ellison (D. Minn.) — State opposition filed 14 August; oral argument on preliminary injunction held 19 August. Next: Preliminary-injunction ruling; no order located as at 21 August.
ALR-0029 Delta Corp v Meta (Bombay High Court) — Interim relief granted 6 August. Next: Listed 27 August 2026.
ALR-0073 TSE Rp 0601617-29.2026.6.00.0000 (Brazil) — Interim removal and disclosure ordered 20 August. Next: Defence submissions and plenary review of the interim decision; no date fixed.
ALR-0072 UMG Recordings v Suno (D. Mass.) — Leave to add §1201 claim granted 18 August. Next: Amended pleading and Suno's response; deadlines require docket extraction.
Autumn and beyond
ALR-0090 Ireland — EU Cloud and AI Development Act consultation — Opened 14 August. Next: Submissions close 11 September 2026.
ALR-0079 Khaira v Charda Punjab and Meta (Chandigarh) — 72-hour takedown ordered 12 August. Next: 14 September 2026 for compliance, written statement and response to the interim application, as reported.
ALR-0078 Khushi Kapoor v Ashok Kumar (Delhi High Court) — Interim injunction 12 August. Next: 17 September 2026 before the Joint Registrar.
ALR-0084 Scale AI v United States (Ct. Fed. Cl.) — Protest rejected 18 August; opinion not obtained. Next: 21 September 2026, when the 20 May oral-argument transcript comes off restriction; any Federal Circuit appeal.
ALR-0017 UK Legal Services Advisory AI Growth Lab — Applications open. Next: Applications close 23:59 on 27 September 2026; decisions expected end of October.
ALR-0051 NAIC AI Risk Evaluation Supplement pilot — 13 August meeting now confirmed by official summary. Next: Further supplement iteration expected end of August; 12-state pilot continues through 30 September 2026.
ALR-0053 South Australia Royal Commission into AI — Announced 11 August; Terms of Reference outstanding. Next: Intended commencement 1 October 2026; final report due by 1 July 2027.
ALR-0091 Colorado ADMT and Chatbot Safety rulemaking — Proposed rules filed 11 August. Next: 5 October 2026 for comments intended to inform revisions; written comments close 26 October 2026; statutes operative 1 January 2027.
ALR-0057 Northern Ireland Draft AI Strategy — Consultation opened 12 August. Next: Closes 17:00 on 7 October 2026.
ALR-0058 NIST National Vulnerability Database AI RFI — RFI opened 12 August. Next: Comments close 23:59 ET on 13 October 2026.
ALR-0020 Ireland — High Court Practice Direction HC 142 — Issued 29 July 2026. Next: Takes effect 1 September 2026.
ALR-0030 Tabu personality-rights proceedings (Delhi) — Interim protection reportedly granted 10 August. Next: Signed order required; reported discovery return 7 December 2026.
ALR-0070 Vietnam Decision No. 33/2026/QĐ-TTg — In force 15 August. Next: Transitional compliance deadlines of 1 March 2027 and 1 September 2027 for pre-existing listed systems; implementation guidance awaited.
Awaiting a trigger — no date fixed
ALR-0033 TSE — Flávio Bolsonaro AI/deepfake proceeding (Brazil) — 13 August meeting postponed; see Correction. Next: Rescheduled internal meeting and formal merits judgment; no court-confirmed date.
ALR-0080 In re OpenAI Copyright Litigation (S.D.N.Y.) — Sanctions application opposed 13 and 14 August. Next: Court's determination; no ruling date located.
ALR-0074 Babylon Bee v Castillo (D.N.M.) — Complaint filed 11 August. Next: Defendants' response and any preliminary-injunction proceedings.
ALR-0035 HateAid / Meta smart glasses (Germany) — Complaint received by ZIT 12 August; no new prosecutorial step located during this reporting period. Next: ZIT decision whether to open a substantive investigation.
ALR-0093 Italy — Atto del Governo No. 421 — Opinion deadline expired 16 August. Next: Confirmation of any committee opinion and subsequent Government adoption.
METHODOLOGY & VERIFICATION NOTE
The AI Law Report separates discovery, verification, database enrichment and publication. Developments are identified through daily global scanning of two streams — courts and litigation, and regulators and public authorities — on a rolling 72-hour window with a seven-day backfill. Candidates are then reconciled against the existing record and against primary or authoritative sources before any entry reaches publication.
Counting and jurisdiction methodology
This edition contains 29 entries: 24 developments reported for the first time, 4 updates to matters opened in earlier issues (ALR-0009, ALR-0039, ALR-0051 and ALR-0052) and 1 correction (ALR-0033). Entries in the Developments to Watch section are not counted again, since each already appears either in this edition or in a previous one.
Jurisdictions are counted once at national level. On that basis this edition draws on 11 national jurisdictions: Australia, Brazil, India, Ireland, Italy, the Philippines, Thailand, Türkiye, the United Kingdom, the United States and Vietnam.
Materiality threshold
Discovery is deliberately broad and applies no materiality threshold; publication does not. The underlying record is wider than this Report. A development is published where it materially advances the legal, regulatory, judicial or public-governance record concerning artificial intelligence.
Thirty-two matters captured during this reporting period are retained in the record but held below the publication threshold or held pending primary material. The largest group is public-sector AI adoption and capability activity that raises no legal or regulatory question at this stage: municipal and state deployments and programmes in Zapopan, Olinda, Yucatán, Rio das Ostras, Torres, Buenos Aires, Ajman, Changhua and Beijing; training and workforce initiatives in Canada, New Zealand, Malaysia and Saudi Arabia; an audit court's internal AI deployment portfolio in Minas Gerais; awards, studies and conference activity in Singapore, Nigeria and Hanoi; a Mexican arts AI-literacy programme; and Taiwan's 2027 budget framing. Infrastructure and financing items, including a Scottish AI growth-zone financing package and a Victorian council's data-centre planning advocacy, are likewise held.
Three further matters are held for reasons of verification rather than materiality, and each would ordinarily qualify for publication. A reported Japanese cross-government package addressing malicious AI use could not be matched to any originating government document dated 14 August, and is not reported until it can be. LeDoux v Outliers, Inc., No. 3:24-cv-05808-TMC (W.D. Wash.), in which contemporaneous reporting describes the action being disposed of on 18 August following a ruling that expert evidence containing AI-generated false citations was unreliable, is held pending the order: the case has significant AI-citation history, and the Report will not characterise the holding from reporting alone. An application by X Corp on 14 August to modify an earlier Gujarat High Court direction concerning the SAHYOG portal, made within an existing deepfake public-interest petition, is held because no signed order has been retrieved; nothing should be inferred about whether that direction has been altered. Two United States federal AI procurement and strategy exercises — a Department of Homeland Security commercial solutions opening on procurement modernisation and a Social Security Administration enterprise AI strategy request for information, both issued in the reporting period — are recorded but held below threshold as routine solicitations. Each of these will be reported if it develops.
Verification and source basis
Every entry carries two statements. Verification records how firmly the development is established: Verified where the underlying event and its material terms are established from authoritative material, Partial where the event is strongly corroborated but an element remains outstanding. Source basis records what was actually seen — a signed judgment, an official government instrument, a docket entry, or contemporaneous reporting where no primary document was available.
Of the 29 entries in this edition, 17 are Verified and 12 are published with partial verification, each with its specific gap identified.
Four qualifications carry particular weight this week. The Scale AI merits opinion has not been obtained, and the Report therefore records the outcome of the protest without stating the court's reasoning. The two São Paulo civil actions rest entirely on Brazilian press reporting: no case numbers, complaints or signed decisions have been retrieved, and the reported refusal of emergency relief in one of them should not be read as a considered ruling on the lawfulness of the video. The Otter.AI order's disposition of individual counts is taken from reporting rather than from the order text, which has not been extracted. And the TSE representation at ALR-0073 is confirmed in indexed form only, the rendered decision being behind the court's access control.
Editorial rules applied
Allegations remain allegations. Consultations are not treated as regulatory decisions. Proposed rules are not described as final regulations. Interim and procedural orders are not presented as final merits determinations. A ruling on forum is not a ruling on the merits. Leave to plead a claim is not a finding that the claim is made out. The expiry of a parliamentary scrutiny deadline is not a parliamentary outcome. Where reporting attributes reasoning to a court whose order has not been obtained, the attribution is made to the reporting rather than to the court.
Headline placement reflects procedural maturity as well as subject-matter interest. Corrections are published at the front of the edition rather than at the back.
Coverage this period materially searched the United States and Canada; the United Kingdom and Ireland; EU institutions and European national jurisdictions; India and South Asia; Australia and New Zealand; East and Southeast Asia using Chinese, Japanese, Korean and Vietnamese search terms; the Middle East and Africa including Arabic and Turkish sources; and Latin America and the Caribbean using Portuguese and Spanish sources. Near-real-time primary docket indexing remains weakest in India, Brazil, parts of continental Europe, Africa and Asia; the absence of a development from a jurisdiction in any given week should not be read as evidence that none occurred. Thailand remains a specific primary-source gap in this edition.
Editorial note
The AI Law Report is an independent editorial publication from AI Policies UK bringing together significant judicial, regulatory and public-authority developments involving artificial intelligence. It reports developments for information and educational purposes and does not provide legal advice or predict legal outcomes. Where possible, reporting is verified against judgments, filings, regulatory publications and other primary sources. Where primary documentation could not be located, this is stated in the relevant entry.
Next edition: Friday 28 August 2026.