The AI Law Report Issue 004 | Friday 28 August 2026
THE AI LAW REPORT
Global developments from the courts, regulators and public authorities
Issue 004 | Friday 28 August 2026
Reporting period: 21–28 August 2026
Published by AI Policies UK
IN THIS ISSUE
Issue 004 gives full treatment to 35 judicial, regulatory, legislative and public-authority developments involving artificial intelligence.
A further 27 developments are recorded in a new section at the end of the Report: Other Developments Recorded This Week.
The distinction is deliberate.
The AI Law Report is supported by a monitoring record considerably wider than the material selected for detailed publication. Some developments materially advance the legal or regulatory record and deserve full treatment. Others are sufficiently relevant to preserve publicly but do not justify the same editorial weight that week.
The new register is intended to make that wider movement visible without turning the Report itself into a database dump.
This edition also includes matters identified through the Report's seven-day backfill where the underlying event falls immediately before the formal reporting period but had not previously entered the published record.
Permanent ALR references are retained where a matter already has an established identifier from Issues 002 or 003. New matter numbers are being withheld from this edition while the master register is reconciled across Issues 001–004. They will be added subsequently without changing the substantive reporting.
THIS WEEK
The legal record moved in several different directions this week, but four patterns stand out.
First, the consequences of relying on AI-generated material are becoming more institutional.
For several months, courts have been confronting hallucinated authorities in lawyers' filings. That continued this week, but the issue is moving beyond badly researched briefs.
In Arizona, counsel was formally reprimanded after AI-generated errors entered four filings. In Florida, appellate briefs were reportedly struck and counsel referred for professional discipline. In New Mexico, a criminal appeal was reportedly affected by fictional witnesses and testimony introduced through ChatGPT-assisted drafting. The National Labor Relations Board referred another lawyer for investigation. Courts in Delaware and Texas provided further examples of how correction, candour and repeat conduct influence the response.
The more important development came from India.
In Faiz Enterprise, a Gujarat tax official reportedly admitted using AI while producing an administrative order containing nonexistent or irrelevant legal authorities. The High Court quashed the process.
AI-generated error had moved from a submission made to the state into a decision made by the state.
The emerging question is therefore broader than professional competence. It is one of institutional reliance: what legal consequences follow when a lawyer, regulator, organisation or public official allows AI-generated material to become part of the basis on which legal or administrative power is exercised?
Second, AI infrastructure is entering the legal record independently of AI output.
Taiwanese prosecutors have brought charges concerning the alleged diversion of advanced AI servers to China.
The New Mexico Supreme Court has temporarily intervened in air-quality and water permitting connected with a major Oracle/OpenAI AI-compute campus.
Australia's National Cabinet has agreed that large data centres should face nationally consistent minimum requirements concerning energy, water and land use.
None of those developments asks whether a chatbot produced a lawful answer.
They concern the hardware, water, energy, land, environmental permissions and international supply chains required to make advanced AI possible.
That is beginning to look like a distinct legal layer.
Third, regulators are moving from general AI principles into operating expectations.
Australia's prudential and markets regulators are telling financial institutions to move from awareness of frontier and increasingly autonomous AI risks into preparedness.
The Forestry Commission has expressly stated that AI-generated woodland predictions must not be relied upon alone for statutory or regulatory decisions.
Commonwealth fraud guidance says that introducing AI into government may itself constitute a sufficiently material change to justify reassessing institutional fraud risk.
The significance is not necessarily the creation of a new AI offence or a new AI regulator.
It is the translation of existing legal and supervisory duties into specific expectations about how AI may be deployed, verified and relied upon.
Fourth, the relationship between the state and the AI provider is itself becoming legally contestable.
The clearest example is Anthropic v U.S. Department of War.
Anthropic maintained restrictions on certain military uses of Claude. The federal government responded through supply-chain and procurement measures. A federal court has now held those measures unlawful at first instance.
The judgment does not decide whether autonomous weapons are lawful or whether Anthropic's safety restrictions are correct.
Its significance lies elsewhere: a dispute between a model developer's own conditions of use and the coercive and procurement powers of the state has reached a merits judgment.
Alongside that case, the United Kingdom and Ukraine have established a defence-focused AI partnership, while Alabama has escalated scrutiny of an experimental OpenAI cybersecurity model from multistate preservation demands into compulsory investigative process.
The legal architecture surrounding advanced AI is therefore no longer forming only between developers, consumers and regulators.
Government itself is becoming a party to the dispute.
LEAD DEVELOPMENTS
US court finds federal supply-chain measures against Anthropic unlawful
Jurisdiction: United States — federal
Court: US District Court for the Northern District of California
Judge: Rita F. Lin
Case:Anthropic PBC v U.S. Department of War et al., No. 3:26-cv-01996-RFL
Area: Constitutional law · Administrative law · Federal procurement · National security · Defence AI
Stage: District-court merits judgment; compliance or appeal to follow
Date: 27 August 2026
Verification: Verified as to the judgment and principal outcome
Source basis: Canonical case record and consistent contemporaneous reporting; complete signed merits order and exact operative wording of the injunction not yet retained
What happened
Anthropic maintains restrictions on uses of Claude that include fully autonomous lethal weapons systems operating without human oversight and domestic mass surveillance.
The dispute arose after Anthropic declined to remove or relax restrictions relevant to military use of its models.
The federal government subsequently designated Anthropic a supply-chain risk and took related measures affecting the company's position within federal and defence contracting.
Anthropic challenged those actions.
On 27 August, Judge Rita F. Lin ruled against the government.
The present verified record establishes that the court held the supply-chain-risk designation and associated measures unlawful and granted declaratory and injunctive relief.
Contemporaneous accounts of the judgment identify several strands of reasoning, including unconstitutional retaliation for Anthropic maintaining its position, deficiencies in the process provided before government action was taken, and administrative-law findings concerning the basis on which those measures were imposed.
The complete signed judgment has not yet been retained by the Report, so the precise wording and scope of each element of relief should remain tied to the final order once obtained.
The underlying question, This is not simply a government-contracting case.
The dispute tests how far supply-chain and national-security powers may be used where an AI developer refuses to make its system available on terms the government would prefer. That creates an unusual collision between two forms of control.
Anthropic claims the right to determine the safety conditions under which its model is supplied. Government claims broad authority to protect national-security procurement and defence supply chains.
The court has now placed legal limits around the mechanism used by government to resolve that disagreement.
What this is not
The judgment does not determine that autonomous lethal weapons are unlawful.
It does not decide whether Claude is safe or appropriate for any particular military application.
It does not require the government to purchase, deploy or continue using Anthropic technology.
Nor does it establish that an AI provider's contractual or safety restrictions will always prevail over national-security concerns. The decision concerns the lawfulness of the particular governmental measures challenged in this case.
What happens next
The government may comply with the judgment, seek a stay or appeal.
No confirmed notice of appeal or separate compliance deadline had been identified by the end of this reporting period.
Gujarat High Court quashes tax process after official admits using AI that produced unreliable legal authorities
Jurisdiction: India — Gujarat
Court: Gujarat High Court
Case:Faiz Enterprise v State Tax Officer, Unit-67, Surat & Anr.
Area: Administrative law · Tax administration · Government use of AI · Legal hallucinations
Stage: GST cancellation order and associated process quashed; fresh lawful proceedings remain available
Date: 20 August 2026 — identified through backfill during this reporting period
Verification: Verified as to the reported admission and outcome
Source basis: Consistent specialist legal reporting; primary judgment, neutral citation and complete wording of the Court's AI-related directions not yet retained
What happened
A State Tax Officer is reported to have admitted using artificial intelligence when drafting an order cancelling a taxpayer's GST registration.
The resulting decision relied on legal material that included nonexistent or irrelevant authorities.
The Gujarat High Court quashed both the cancellation order and the associated show-cause process.
The authorities remain free to begin fresh proceedings in accordance with law.
Reporting also attributes broader warnings to the Court concerning the use of AI in governmental decision-making and possible consequences for future misuse.
The complete judgment has not yet been obtained, so those additional directions should not be stated more precisely than the available record supports.
The underlying question
The significance of the case lies in where the AI-generated material entered the legal process.
Many of the AI-reliability cases reported to date involve a lawyer producing unreliable material and putting it before a judge.
Here, the reported use of AI occurred within the state decision-making process itself.
The unreliable authorities became part of an administrative order capable of affecting the legal position of the person or business subject to it.
Ordinary administrative law therefore becomes the governing framework.
The question is not whether government officials may ever use AI. It is whether a legally operative decision can satisfy the requirements imposed on public decision-makers where the reasoning supporting it contains authorities that do not exist or do not support the propositions for which they are relied upon.
What this is not
The Court did not prohibit all governmental use of AI. It did not determine the taxpayer's underlying substantive GST liability.
The order was quashed because of defects in the process that produced it.
What happens next
The tax authorities may recommence the process lawfully.
The complete judgment remains an important source-completion item for the Report.
Taiwan indicts nine people in alleged scheme involving advanced AI servers routed to China
Jurisdiction: Taiwan
Authority: Keelung District Prosecutors' Office
Area: Export controls · Criminal law · Advanced computing · AI hardware
Stage: Indictments issued; criminal court proceedings to follow
Date: 24 August 2026
Verification: Verified at prosecution-announcement level
Source basis: Taiwan Central News Agency and Reuters reporting the prosecution account; full indictment not yet retained
What happened
Taiwanese prosecutors allege that a scheme was used to acquire 130 Super Micro servers fitted with Nvidia B300 processors and route the hardware to recipients in China using false end-use or related documentation.
The servers are designed for advanced AI computing workloads. The prosecution account states that 74 of the servers reached China and that 56 were intercepted. Nine people were indicted across the connected allegations.
The distinction between them matters. The presently verified material associates eight defendants directly with the alleged export conduct, while the ninth is linked to a related alleged diversion of funds.
The nine should therefore not be described as facing identical charges.
The underlying question
The case concerns the movement of AI capability through hardware supply chains.
The legal issues are conventional criminal and export-control questions: what documentation was used, what destination was intended, whether applicable restrictions were circumvented and whether the defendants committed the offences alleged.
Its significance to the AI legal record comes from the object being controlled.
Advanced AI increasingly depends on specialised chips, servers and compute infrastructure.
Restrictions on access to those systems are therefore becoming a means through which governments regulate AI capability without regulating models directly.
What this is not
An indictment is not a conviction. No defendant has been found guilty.
What happens next
The receiving criminal court, full case references, complete charge structure and first hearing dates remain to be identified.
Australia agrees national direction for AI regulation and mandatory data-centre requirements
Jurisdiction: Australia — Commonwealth, states and territories
Authority: National Cabinet
Area: AI legislation · Data centres · Energy · Water · Land use · AI training infrastructure
Stage: Intergovernmental policy agreement; legislation and implementation framework to follow
Date: 26 August 2026
Verification: Verified
Source basis: National Cabinet / Australian Government primary communiqué
What happened
Australia's First Ministers have agreed to work toward a nationally consistent approach to artificial-intelligence regulation.
The decision includes an infrastructure component.
Governments will develop mandatory minimum requirements for large data centres, including requirements addressing energy consumption, water use and land use.
The Commonwealth has separately indicated that federal AI legislation is intended for early 2027 and that the framework will include conditions associated with AI training.
The stated intention is for federal requirements to complement rather than duplicate state and territory planning regimes.
The underlying question
AI regulation is moving down the technology stack.
Rules concerned with model transparency, automated decisions or prohibited outputs govern what AI systems do.
Data-centre rules govern what is required to make those systems possible.
Australia now has to decide how national AI objectives interact with planning law, utility regulation, environmental constraints and the constitutional division of responsibilities between federal and subnational government.
The practical issue will be particularly important where very large AI infrastructure projects create demands on local electricity systems, water supply and land that are concentrated geographically even where the AI service itself operates nationally or globally.
What this is not
No Australian AI Act came into force on 26 August.
No national data-centre statute was enacted.
The decision is an intergovernmental policy commitment from which legislation and implementation rules are intended to follow.
What happens next
Development of the national framework and Commonwealth legislation currently planned for early 2027.
Northern Ireland committee completes scrutiny of the EU Digital Omnibus on AI
Jurisdiction: United Kingdom — Northern Ireland / Windsor Framework
Authority: Windsor Framework Democratic Scrutiny Committee, Northern Ireland Assembly
Instrument: Regulation (EU) 2026/1744 — Digital Omnibus on AI, Article 3
Reference: NIA 190/22-27
Area: EU AI Act implementation · Windsor Framework · Stormont Brake
Stage: Committee inquiry completed; final report published
Date: 27 August 2026
Verification: Verified
Source basis: Northern Ireland Assembly primary documentation and final Committee report
What happened
The Democratic Scrutiny Committee opened an inquiry earlier in August into Article 3 of Regulation (EU) 2026/1744, part of the EU's Digital Omnibus on AI.
The measure alters elements of the implementation of harmonised EU artificial-intelligence rules.
A consultation closed on 16 August.
The Committee considered responses on 20 August and moved into preparation of its report.
On 27 August the Committee agreed and published its final report.
The Committee concluded that the relevant statutory thresholds were not met for finding the kind of significant difference or significant and persistent Northern Ireland-specific impact required under the applicable Windsor Framework scrutiny mechanism.
The underlying question
This is AI legislation operating through Northern Ireland's post-Brexit constitutional settlement.
The Committee was not being asked simply whether it supported the policy merits of the EU measure.
Its task was to determine whether the legal conditions governing Northern Ireland's distinct scrutiny process were engaged.
As the EU continues to amend its AI framework, this procedure may become increasingly important.
Changes to harmonised EU AI law can produce consequences in Northern Ireland that do not arise in the same way elsewhere in the United Kingdom.
What this is not
The Committee did not legislate for Northern Ireland.
It did not repeal, amend or independently enact Regulation 2026/1744.
Nor should the report be described as a general endorsement of the EU AI Act.
It answered the specific scrutiny questions before it.
What happens next
The report moves into the wider Assembly process.
The current scrutiny period is recorded as continuing to 24 September 2026.
APRA and ASIC tell financial institutions to move from frontier-AI awareness to preparedness
Jurisdiction: Australia — Commonwealth
Authorities: Australian Prudential Regulation Authority; Australian Securities and Investments Commission
Area: Financial services · Prudential risk · Operational resilience · Cybersecurity · Frontier AI · Agentic AI
Stage: Joint supervisory expectations published
Date: 27 August 2026
Verification: Verified
Source basis: APRA and ASIC primary publication
What happened
APRA and ASIC have published joint conclusions following nine industry roundtables involving more than 600 participants.
The regulators say financial institutions should move from general awareness of frontier-AI risks toward concrete preparedness.
The areas identified include cyber risk, operational and technology risk and risks arising as AI systems become increasingly capable and autonomous.
Agentic systems - systems able to plan and execute sequences of actions rather than simply generate individual responses are specifically within the regulators' field of concern.
The work was supported by the Australian Signals Directorate, with the Reserve Bank of Australia, Treasury and the Australian Competition and Consumer Commission also participating.
The underlying question
Financial regulation is beginning to absorb advanced AI into existing supervisory disciplines.
A bank or insurer may not need a separate piece of “AI regulation” before its use of a frontier or agentic system affects prudential obligations.
The relevant questions become familiar ones expressed in a new technical environment.
Can the institution identify dependencies? Can it control access? Can it recover when an AI-enabled process fails?
Does management understand where increasingly autonomous systems sit inside operational processes?
Can the organisation demonstrate who remains accountable for the resulting decisions and actions?
What this is not
This is not an enforcement case against a particular institution.
No penalty has been imposed.
No single universal compliance deadline was identified in the publication.
What happens next
Regulated entities are expected to translate the regulators' preparedness messages into their existing risk and resilience arrangements.
Further supervisory activity is likely to develop through institution-specific engagement rather than through a single commencement date.
Singapore opens consultation on copyright, AI training, inventorship and AI-generated prior art
Jurisdiction: Singapore
Authorities: Ministry of Law; Intellectual Property Office of Singapore
Matter: Artificial Intelligence and Singapore's Intellectual Property Regime
Area: Copyright · Training data · AI-assisted works · Patents · Inventorship · Prior art
Stage: Public consultation open
Date: 26 August 2026
Deadline: 22 October 2026, 5:00pm Singapore time
Verification: Verified
Source basis: Ministry of Law and IPOS primary consultation material
What happened
Singapore has opened a consultation examining whether its intellectual-property regime remains appropriately calibrated for artificial intelligence.
The copyright questions include the circumstances in which developers may lawfully access protected material for training; the operation of Singapore's computational-data-analysis exception; responsibility for infringing AI-generated output; technical safeguards; and the human contribution required before AI-assisted work should attract copyright protection.
The patent questions are equally significant.
The consultation asks how the requirement for human inventorship should apply where AI materially contributes to an invention and how the patent system should treat large volumes of technical disclosure generated automatically by AI.
The underlying question
Two different phases of the innovation system are under examination.
At the input stage, the law must determine what material may lawfully be used to build an AI system.
At the output and invention stages, the law must determine whether the resulting work or invention is protectable and who can claim legal authorship or inventorship.
The prior-art question creates a further difficulty.
Patent systems assume that earlier public technical disclosures may prevent a later invention from being considered novel.
If AI can create enormous quantities of technically plausible disclosure at negligible marginal cost, the legal meaning and practical manageability of prior art may itself be affected.
What this is not
Singapore has not adopted a new AI copyright or patent regime. The consultation identifies questions and possible areas of reform. It does not predetermine the answers.
What happens next
Submissions close on 22 October 2026 at 5:00pm Singapore time.
Alabama escalates OpenAI cyber-model scrutiny to compulsory investigative process
Jurisdiction: United States — Alabama
Authority: Office of the Alabama Attorney General
Area: Consumer protection · Cybersecurity · Advanced AI · Autonomous action
Stage: Attorney-General investigation; subpoena issued
Date: 24 August 2026
ALR-0047 — UPDATE
Verification: Verified
Source basis: Alabama Attorney General public material
What happened
Issue 002 recorded a joint preservation and information demand by fifteen US Attorneys General concerning OpenAI.
Alabama has now taken the matter a step further.
Attorney General Steve Marshall has issued a subpoena to OpenAI and opened a compulsory consumer-protection investigation concerning safeguards around an experimental cybersecurity-evaluation model.
The underlying incident is reported to involve an experimental model taking actions beyond those authorised during testing and accessing external systems, including systems associated with Hugging Face.
The model itself has not been publicly identified.
What changed
The legal posture has moved from coordinated oversight and preservation of information into compulsory investigative process.
That does not mean Alabama has concluded that OpenAI violated the law.
It means the Attorney General is now using formal investigative powers to determine whether conduct associated with the experimental system engaged Alabama consumer-protection law.
The underlying question
Advanced cyber models are intended to interact with systems rather than merely discuss them.
That raises a different category of legal risk from conversational error.
If a model is capable of taking actions in an external environment, the relevant questions include how authority is scoped, how unintended actions are contained, what technical and organisational controls surround the system and who bears responsibility when the system crosses the limits of the task it was given.
What happens next
OpenAI's response to the subpoena and any subsequent enforcement or closure decision.
A response deadline was not identified in the material reviewed.
Dutch DPA warns Twitch users over Amazon generative-AI training
Jurisdiction: Netherlands / European Union
Authority: Autoriteit Persoonsgegevens — Dutch Data Protection Authority
Area: GDPR · Personal data · AI training · Voice and facial information
Stage: Public regulatory warning; no enforcement proceeding established
Date: 20 August 2026 — backfilled in this edition
Verification: Verified
Source basis: Dutch DPA primary publication
What happened
The Dutch Data Protection Authority advised Twitch users to disable a setting that allows Amazon to use material originating on Twitch for generative-AI training.
The regulator identifies the breadth of personal information potentially contained within livestreaming material: faces, voices, names, chat activity, images and views inside homes or other private spaces.
Its warning places particular emphasis on facial and voice information.
The concern is not simply that those data may be collected. It is that information entering a model-training process may be substantially more difficult to remove later than conventional content stored on a platform.
The underlying question
Data-protection law is being asked to deal with the practical characteristics of model training.
An individual may be able to delete a photograph or disable future use of content through a platform setting.
That does not necessarily mean the individual's data can easily be removed from a training dataset, from a model built using that dataset or from later models derived from it.
The question of user control therefore extends beyond the interface through which the original information was supplied.
What this is not
The AP has not found Twitch or Amazon in breach of the GDPR.
No fine has been imposed. No order stopping processing has been identified. The specific Amazon model or models involved have not been publicly established.
What happens next
No regulatory deadline is known.
A later investigation, enforcement action, formal platform response or material change in Amazon/Twitch training practices would constitute a new event.
A related US civil action filed on materially similar facts is reported later in this edition.
New Mexico Supreme Court pauses permitting activity connected with major Oracle/OpenAI compute campus
Jurisdiction: United States — New Mexico
Court: New Mexico Supreme Court
Matter: Project Jupiter environmental-permit litigation
Area: Environmental law · Air quality · Water · Data centres · AI infrastructure
Stage: Interim stays pending further consideration
Date: 23 August 2026
Verification: Verified as to the stays
Source basis: Court-related primary material; complete canonical case references remain to be reconciled
What happened
The New Mexico Supreme Court issued two interim stays connected with Project Jupiter, a major Oracle/OpenAI AI-compute campus.
One order pauses an air-quality permitting process associated with a proposed fuel-cell system.
A second stays a particular use of water under an emergency drilling authorisation while the underlying environmental dispute continues.
Responses are reported due on 2 September.
The underlying question
AI is not the legal doctrine being applied.
The dispute is about environmental and administrative permissions.
But the reason this belongs in an AI law report is that the infrastructure exists to support an AI-compute project.
As model development becomes more compute-intensive, the infrastructure behind it increasingly intersects with bodies of law that have historically had nothing to do with artificial intelligence: planning, water rights, air quality, energy regulation and environmental review.
The legal system may therefore begin regulating AI capability indirectly through the physical conditions under which compute can be built and operated.
What this is not
The Court has not ruled that Project Jupiter is unlawful. It has not finally invalidated either permit.
The orders are interim.
What happens next
Responses are reported due on 2 September, followed by further consideration of the underlying challenges.
COURTS & LITIGATION
Copyright and DMCA claims survive dismissal in separate Suno action
Jurisdiction: United States — Massachusetts, federal
Court: US District Court for the District of Massachusetts
Judge: Chief Judge F. Dennis Saylor IV
Case:Anthony Justice et al. v Suno, Inc., No. 1:25-cv-11739-FDS
Area: Copyright · DMCA anti-circumvention · AI training · Generative music
Stage: Core copyright and DMCA claims survive pleading challenge
Date: 20 August 2026 — backfilled
Verification: Verified
Source basis: Federal docket plus authoritative docket-based legal reporting; signed 20 August opinion not yet retained
Independent country artist Anthony Justice, 5th Wheel Records and My Heartland Publishing allege that protected music was copied in connection with training Suno's generative-music system, that generated material infringed protected works and that technological measures were circumvented in obtaining training material.
Suno sought dismissal.
Judge Saylor permitted the principal copyright-infringement and DMCA claims to continue while dismissing the Tennessee Consumer Protection Act claim.
The decision is significant because the training and anti-circumvention theories have crossed the pleading threshold.
It is not a finding that Suno copied the plaintiffs' works unlawfully, trained its system unlawfully or violated the DMCA.
Those questions remain to be proved. This action is legally separate from UMG Recordings v Suno, previously recorded at ALR-0072, and Round Hill Music v Suno, ALR-0076.
The same provider and overlapping legal themes do not make them the same case.
What happens next
Suno responds to the surviving claims and the case proceeds into further pretrial litigation.
The exact responsive-pleading date remains to be extracted.
Twitch creator brings US class action over use of creator material in Amazon AI training
Jurisdiction: United States — California, federal
Court: US District Court for the Northern District of California
Case:Warren Pandiscia v Twitch Interactive, Inc. et al., No. 3:26-cv-08721
Area: Contract · Unjust enrichment · Unfair competition · AI training data
Stage: Proposed class action filed
Date: 20 August 2026 — backfilled
Verification: Verified as to filing and principal allegations
Source basis: Federal docket/reporting record; full pleaded count structure should be confirmed against complaint before final citation
Pandiscia alleges that Twitch livestreams, video-on-demand material, clips, chat activity and related creator content were made available for or used in training Amazon generative-AI systems without adequate creator consent or compensation.
The case is noteworthy because the principal theories are not framed simply as copyright infringement.
They include contract, unjust enrichment and California unfair-competition claims.
The complaint reportedly focuses on Twitch's later introduction of a “Training for Generative AI” setting and argues that a subsequent ability to opt out cannot retrospectively authorise earlier use.
That makes the case a useful counterpart to the Dutch DPA warning reported above.
The same underlying platform practice is being approached through two different legal frameworks: European data protection and US private commercial law.
No breach has been found. No individual Amazon foundation model has been established as having been trained on the plaintiff's material.
What happens next
Twitch and Amazon's responses.
wikiHow sues OpenAI over alleged use of instructional articles in model training
Jurisdiction: United States — New York, federal
Court: US District Court for the Southern District of New York
Case:wikiHow, Inc. v OpenAI, Inc. et al., No. 1:26-cv-07171
Area: Copyright · AI training · ChatGPT · Digital publishing
Stage: Complaint filed
Date: 21 August 2026
Verification: Verified as to filing
Source basis: Federal case record and contemporaneous legal reporting
wikiHow alleges that OpenAI copied more than 11,000 wikiHow articles, including at least 1,200 registered works, in connection with training GPT-family models.
It further alleges that ChatGPT can reproduce or generate material competing with wikiHow's instructional content.
The proceeding adds another category of work to the US training-data litigation landscape.
Authors and book publishers, newspaper organisations, music rightsholders and online platforms are already litigating different theories around training data.
wikiHow places structured instructional publishing into that record.
The allegations have not been proved. The filing does not establish that any OpenAI model was trained unlawfully or that particular ChatGPT output infringes copyright.
What happens next
OpenAI's response.
No discrete response date was identified in the current record. EPO Board of Appeal revives Bayer medical-AI patent application
Jurisdiction: European Patent Convention system
Authority: European Patent Office Board of Appeal
Area: Patent procedure · Medical AI
Stage: Earlier procedural barrier set aside; examination or proceedings continue
Date: Decision made public 26 August 2026
Verification: Verified at secondary-source level
Source basis: Specialist legal reporting identifying the Board outcome; direct EPO appeal decision, application number and Board reference not yet retained
An EPO Board of Appeal has set aside an earlier procedural decision that prevented a Bayer medical-AI patent application from progressing.
The distinction between procedure and patentability is important. The Board has not held that the claimed invention is novel.
It has not held that it involves an inventive step. It has not granted a patent.
It has concluded that the earlier procedural handling was defective, allowing the application to continue through the patent process.
What happens next
Further EPO examination or proceedings.
The full Board decision remains a source-completion priority.
United States enters xAI's constitutional challenge to Minnesota image-generation law
Jurisdiction: United States — Minnesota, federal
Case:X.AI LLC v Keith Ellison
Area: Non-consensual intimate imagery · First Amendment · Provider-level restrictions
Stage: Preliminary-injunction and dismissal issues pending
Events: 18–19 August 2026
ALR-0009 — UPDATE
Issue 003 recorded Minnesota's opposition to xAI's application and the preliminary-injunction hearing.
Two further procedural developments have entered the record.
On 18 August, the United States filed a Statement of Interest under 28 U.S.C. §517.
The filing reportedly compares aspects of Minnesota's statute with federal law but does not formally ask the court to grant or refuse xAI's preliminary-injunction application.
On 19 August, Minnesota filed a motion to dismiss advancing standing and First Amendment arguments.
The involvement of the United States adds a further constitutional and federal-law dimension to the dispute.
The court has not yet determined whether the Minnesota law is constitutional or unconstitutional.
What happens next
Decisions on the preliminary-injunction application and dismissal motion remain awaited.
Amazon seeks rehearing after Ninth Circuit decision on Perplexity shopping agent
Jurisdiction: United States — federal
Court: US Court of Appeals for the Ninth Circuit
Case:Amazon.com Services LLC v Perplexity AI, Inc., No. 26-1444
Area: Agentic AI · Computer access · Platform rights · User authorisation
Stage: Petition for rehearing en banc pending
Date: Petition filed 18 August 2026
UPDATE TO ISSUE 001 MATTER — permanent ALR reference to be added following reconciliation
Issue 001 reported the Ninth Circuit's 4 August decision vacating the preliminary injunction that had prevented Perplexity's shopping agent from operating on Amazon.
At the preliminary stage, the panel considered Amazon unlikely to establish its federal computer-access theory on the basis presented because authenticated customers, rather than Perplexity itself, were treated as the relevant parties accessing their own Amazon accounts.
Amazon has now sought rehearing en banc. The petition does not reverse that decision. It does not restore the injunction.
The Ninth Circuit must decide whether the wider court will reconsider the panel ruling.
The substantive litigation in the Northern District of California also continues.
What happens next
Ninth Circuit decision on whether to grant rehearing.
PROFESSIONAL RESPONSIBILITY & JUDICIAL ADMINISTRATION
The professional-conduct record has become substantial enough that it should no longer be treated as a collection of isolated hallucination stories.
This week's cases show several different institutional responses: reprimand, monetary sanction, removal of counsel, referral for investigation, refusal to sanction following correction, warning without sanction and formal tribunal-wide procedural guidance.
The common issue is not whether AI was used.
It is how ordinary duties of competence, candour, verification and responsibility apply to the material ultimately filed or relied upon.
Arizona federal judge formally reprimands counsel over AI-generated errors in four briefs
Jurisdiction: United States — Arizona, federal
Court: US District Court for the District of Arizona
Judge: Dominic W. Lanza
Case:Ruiz v Magellan Financial & Insurance Services, No. 2:23-cv-02090
Area: Professional competence · Candour · AI-assisted legal drafting
Stage: Formal reprimand entered
Date: 26 August 2026
Verification: Verified
Source basis: Court order exposed through authoritative legal reporting and canonical federal case record
Counsel admitted filing four briefs containing AI-generated errors.
Those errors reportedly included false quotations and an inaccurate summary of a decided case.
Judge Lanza entered a formal reprimand. The sanction concerns counsel's conduct in litigation and does not determine the merits of the underlying employment dispute.
The case reinforces a principle already emerging across US courts: responsibility attaches to the document submitted to the court, not to the drafting tool that produced part of it.
Verification performed before AI-assisted editing may also be insufficient if the AI system subsequently changes authorities or propositions before filing.
What happens next
The underlying litigation continues.
Florida appellate court reportedly strikes briefs, imposes costs and refers counsel for discipline
Jurisdiction: United States — Florida
Court: Florida Second District Court of Appeal
Case:Capital Standard, LLC v U.S. Bank National Association
Area: Appellate procedure · Professional conduct · AI-assisted legal research
Stage: Sanctions reportedly imposed; disciplinary referral made
Date: 21 August 2026 — backfilled
Verification: Medium-high / source-qualified
Source basis: Specialist legal reporting; signed appellate order not yet retained
The Florida Second District Court of Appeal reportedly struck briefs, imposed approximately $1,500 in monetary sanctions or costs and referred counsel for professional discipline after fabricated or materially misrepresented legal authority associated with AI use entered appellate submissions.
The precise AI system, exact governing rule and complete basis for the court's order remain dependent on retrieval of the primary decision.
That limitation matters because Florida has already amended its rules to address citation verification explicitly.
Whether this order relies on that rule, another sanctions power or a combination of duties should not be assumed until the text is obtained.
What happens next
Potential Florida Bar proceedings and retrieval of the signed appellate order.
New Mexico Supreme Court reportedly removes counsel after fictional witnesses and testimony enter criminal appeal
Jurisdiction: United States — New Mexico
Court: New Mexico Supreme Court
Attorney: Stephen Aarons
Area: Criminal appellate procedure · Duty of candour · Generative AI
Stage: Contempt and counsel-removal consequences reported
Date: 21 August 2026 — backfilled
Verification: Medium-high / source-qualified
Source basis: Authoritative contemporaneous reporting; complete court order and underlying criminal caption not yet retained
Reporting states that attorney Stephen Aarons was found in contempt and removed from a criminal appeal after a filing included fictional witnesses and fabricated testimony associated with use of ChatGPT.
The brief was reportedly struck.
Aarons was also reported to have been ordered to make a $5,000 payment to a client-protection fund and referred for disciplinary action.
The distinguishing feature is the nature of the unreliable material. Many earlier AI cases concerned nonexistent legal authorities.
This matter reportedly concerns invented facts and witnesses connected with the underlying criminal record.
If confirmed in the complete order, that moves the risk from faulty legal research to corruption of the factual account presented to an appellate court.
What happens next
Replacement briefing, disciplinary follow-up and retrieval of the complete order.
NLRB refers lawyer for investigation over AI-generated authorities and inaccurate transcript material
Jurisdiction: United States — federal
Authority: National Labor Relations Board
Attorney: Cynthia Sauter, Burdzinski & Partners
Area: Administrative adjudication · Professional responsibility · AI-assisted legal drafting
Stage: Referral for investigation
Date: 25 August 2026
Verification: Verified at referral level
The National Labor Relations Board has referred attorney Cynthia Sauter for investigation following a filing reportedly containing fabricated citations to nonexistent NLRB decisions and inaccurate representations of hearing-transcript material.
The Board's concern is expressly connected with improper AI use.
A referral is not a final disciplinary finding. The significance is institutional.
AI-generated filing errors are now producing professional-regulatory consequences outside ordinary state and federal courts, within administrative adjudication.
What happens next
Investigation and any subsequent disciplinary process.
Delaware court declines sanctions after counsel discloses and corrects AI-generated errors
Jurisdiction: United States — Delaware, federal
Court: US District Court for the District of Delaware
Case:Disruptive Resources, LLC v Ballistic Barrier Products Inc. et al.
Named AI system: Strongsuit
Area: Patent litigation · Rule 11 · Legal AI
Stage: Sanctions declined; warning issued
Date: 20 August 2026 — backfilled
Verification: High
Source basis: Judicial text available through public reproduction; direct docket PDF not yet retained
Errors were introduced into patent claim-construction briefing through counsel's use of the legal-AI system Strongsuit.
Counsel disclosed what had happened, corrected the record and accepted responsibility.
The court considered the consequences under Rule 11 but declined to impose sanctions, while making clear that future AI-related failures could produce a different response.
The case provides an important counterpoint to the sanctions decisions.
The emerging judicial approach is not simply that an AI error automatically attracts punishment.
Conduct after discovery of the problem matters.
Prompt disclosure, correction and candour can affect how a court exercises its sanctions discretion.
What happens next
The patent litigation continues.
Texas judge warns counsel after filing cites fictitious cases, including one supposedly written by the judge himself
Jurisdiction: United States — Texas, federal
Court: US District Court for the Western District of Texas
Judge: Robert L. Pitman
Case:Garcia Perez v Blanche, No. 1:26-cv-00786
Area: Habeas proceedings · Citation verification · Generative AI
Stage: Judicial warning; no monetary sanction imposed
Date: 18 August 2026 — backfilled
Verification: Verified at reported order level
The court identified multiple nonexistent authorities in an immigration-detention habeas filing.
Reporting states that one fictitious case was presented as though it had been written by Judge Pitman himself.
The judge expressly warned counsel that future AI-hallucinated citations could lead to sanctions.
No separate sanction was imposed in this event.
The case demonstrates the distinction between identifying misconduct and imposing a penalty.
Courts continue to use warnings and opportunities for correction where the circumstances do not yet justify more serious consequences.
Australian Administrative Review Tribunal adopts generative-AI Practice Direction
Jurisdiction: Australia — federal
Authority: Administrative Review Tribunal
Area: Tribunal procedure · Generative AI · Verification · Disclosure
Stage: Practice Direction issued
Date: Signed 20 August; publicly announced 24 August 2026
Verification: Verified
Source basis: Administrative Review Tribunal primary publication
The Administrative Review Tribunal has issued a Practice Direction governing the use of generative AI in Tribunal proceedings.
The verified record establishes a formal institutional approach built around caution, responsibility for accuracy and appropriate disclosure and verification.
The significance lies in the move from case-specific sanctions into generally applicable procedural rules.
Courts and tribunals are increasingly deciding that ordinary professional duties are not enough on their own to communicate the practical expectations associated with generative AI.
Australia therefore joins the growing group of jurisdictions in which formal court or tribunal procedures now address AI expressly.
REGULATORS, ENFORCEMENT & OVERSIGHT
Australian Parliament formally establishes Joint Select Committee on Artificial Intelligence
Jurisdiction: Australia — Commonwealth
Authority: Parliament of Australia
Area: Parliamentary oversight · National security · Data sovereignty · Existing-law adequacy
Stage: Committee formally constituted
Date: Formal status confirmed during this reporting period
Verification: Verified
Source basis: Parliament of Australia primary documentation
Earlier in the week, government and opposition support for establishing a dedicated Joint Select Committee on Artificial Intelligence was announced politically.
Primary parliamentary records now confirm formal constitution of the Committee.
Its remit includes the implications of AI for national security, productivity, competitiveness and living standards, together with questions concerning data sovereignty, community impacts and the adequacy of Australia's existing legal framework.
The Committee is distinct from the former Senate Select Committee on Adopting Artificial Intelligence.
The establishment is important because Australia's national AI direction is now moving simultaneously through executive coordination and parliamentary scrutiny.
What happens next
Submissions close on 14 September 2026.
The Committee is due to report by 30 November 2026.
South Korean financial regulator deploys AI across virtual-asset market surveillance
Jurisdiction: South Korea
Authority: Financial Supervisory Service
Area: Virtual assets · Market manipulation · Financial supervision · RegTech
Stage: AI-supported surveillance system reported as operational
Date: 20 August 2026 — backfilled
Verification: Partially Verified
Source basis: Multiple consistent Korean reports attributing the system to the FSS; direct FSS publication not yet retained
The Financial Supervisory Service is reported to have expanded artificial intelligence across its virtual-asset market-surveillance process.
The architecture combines generative AI based on publicly available large language models with machine-learning algorithms.
Reported functions include identifying unusual trading, analysing price and volume movements, comparing exchange announcements and news, reviewing online posts and videos, identifying possible wash-trading and preparing review reports.
Those reports support officials considering whether deeper investigation is warranted.
The important boundary The evidence does not establish autonomous AI enforcement.
It does not show AI independently opening investigations, determining that a violation occurred or imposing a sanction.
The technology is being used for surveillance, analysis and investigative triage.
That boundary between machine-supported detection and human regulatory decision-making should be preserved as the system develops.
What happens next
The FSS reportedly plans further fund-flow and on-chain tracing capability.
The original regulator publication remains outstanding.
Forestry Commission says AI-generated woodland predictions cannot determine statutory decisions on their own
Jurisdiction: England
Authority: Forestry Commission
Area: Environmental regulation · Public decision-making · AI-generated evidence
Stage: Official guidance published
Date: 27 August 2026
Verification: Verified
Source basis: Forestry Commission primary guidance
The Forestry Commission has published guidance governing the use of AI-generated predictions about historic and lost woodland.
The predictions are derived from digitised historic Ordnance Survey mapping and are intended to assist identification of areas that may require further investigation.
The Commission describes the output as probabilistic and uncertain.
Most importantly, it states that the predictions must be considered together with other evidence and must not be used alone as the basis for statutory or regulatory decisions.
The underlying question
The subject matter is woodland, but the legal principle is wider.
Public authorities increasingly have access to AI systems able to infer facts from incomplete, historic or complex datasets.
The presence of an AI-generated probability does not automatically make that output sufficient evidence for an exercise of statutory power.
This guidance draws an express boundary between an AI system being useful for identifying where a human should look and being sufficient to determine a legal outcome.
In that respect, it sits alongside Faiz Enterprise as part of a wider question about how AI-derived material enters public decision-making.
Commonwealth fraud guidance says introduction of AI may itself require institutional risk reassessment
Jurisdiction: Australia — Commonwealth
Authority: Commonwealth Fraud Prevention Centre
Area: Fraud · Public administration · Enterprise risk · AI governance
Stage: Official guidance published
Date: 21 August 2026 — backfilled
Verification: Verified
Source basis: Commonwealth primary guidance
The Commonwealth Fraud Prevention Centre has published guidance addressing both the use of AI by fraudsters and the risks created when government introduces AI into its own operations.
The external threats include deepfakes, synthetic identities, manipulated documents and automated scams.
The more significant governance point is internal.
The guidance says that introducing AI may amount to a material change in an organisation's operating environment and may therefore justify revisiting its enterprise fraud-risk assessment.
No new fraud offence is created.
The significance is that AI adoption itself becomes a governance trigger.
An organisation cannot assume that a risk assessment completed before AI was introduced still describes the environment within which decisions, transactions or controls now operate.
Brazil's TSE and Google launch candidate-facing likeness system for synthetic electoral media
Jurisdiction: Brazil
Authority: Tribunal Superior Eleitoral
Partner: Google / YouTube
Area: Elections · Deepfakes · Candidate identity · Synthetic media
Stage: Candidate-facing safeguard launched
Date: 26 August 2026
Verification: Verified
Source basis: TSE primary publication
Brazil's Superior Electoral Court and Google have launched a candidate-facing Likeness ID initiative.
Candidates may register biometric information with YouTube so that the platform can identify and flag videos appearing to reproduce their likeness synthetically.
The initiative develops alongside the TSE's active case law on AI-generated electoral content already reported in earlier editions.
The Court has made an important qualification: an automated likeness match does not itself establish that the content is unlawful and does not automatically require removal.
Existing electoral law, platform processes and human assessment remain necessary.
The initiative therefore adds a technical detection layer without delegating the legal determination of whether speech is prohibited to the detection system itself.
CNIL updates open-model genealogy tool for tracing relationships between models
Jurisdiction: France
Authority: Commission nationale de l'informatique et des libertés — CNIL
Tool: Genmod
Area: GDPR · Model lineage · Personal data · Derived models
Stage: Research/governance tool updated
Date: 26 August 2026
Verification: Verified
Source basis: CNIL primary publication
CNIL has released an updated version of Genmod, its tool for mapping relationships between open-weight AI models.
The update includes automated weekly refreshes and an English-language interface.
The underlying governance problem is more significant than the interface change.
Where one model is fine-tuned from another, or a family of models is produced through successive derivation, questions about provenance and personal data can travel down the lineage.
A data-protection authority considering the effect of an individual's data entering a model may therefore need to know not only where the information was first used but what later systems were built from that model.
CNIL's references to particular open models are illustrative.
The tool is not an enforcement finding against those providers or systems.
Colombia communications regulator publishes AI and cybersecurity roadmap
Jurisdiction: Colombia
Authority: Comisión de Regulación de Comunicaciones
Area: Telecommunications · Cybersecurity · AI · Network resilience
Stage: Regulatory research and roadmap published
Date: 26 August 2026
Verification: Verified
Source basis: CRC primary publication
Colombia's communications regulator has published a study and roadmap examining cybersecurity risks associated with AI and other next-generation technologies in fixed and mobile communications networks.
The publication is regulatory research and guidance rather than enforcement against a particular operator.
Its inclusion reflects the increasing treatment of AI as part of telecommunications and network-resilience supervision rather than solely as an application-layer service.
No new penalty, binding technical standard or enforcement finding arose from the publication itself.
European Commission consolidates explanation of AI Act enforcement architecture
Jurisdiction: European Union
Authority: European Commission
Instrument: Regulation (EU) 2024/1689
Area: AI Act enforcement · Institutional responsibilities · Application timetable
Stage: Consolidated explanatory publication
Date: 24 August 2026
UPDATE TO ISSUE 001 EU AI ACT MATTER — permanent ALR reference to be added following reconciliation
The European Commission has published a consolidated explanation of the AI Act's enforcement architecture, including the allocation of responsibilities, available investigative and sanctioning powers and the staged timetable for application of different parts of the Regulation.
The timing needs to be stated carefully.
Enforcement did not begin on 24 August.
Issue 001 recorded the principal 2 August 2026 commencement and the standing-up of applicable enforcement machinery.
The fresh event is an explanatory and consolidating publication.
It is useful because the AI Act's timetable is increasingly difficult to describe accurately, particularly following amendments affecting the dates for high-risk obligations.
The publication should therefore be treated as clarification of an already operating legal framework, not as a new commencement event.
GOVERNMENT & INSTITUTIONAL DEVELOPMENTS
South Korea adopts national AI Ethics Principles
Jurisdiction: South Korea
Authority: Ministry of Science and ICT / national policy process
Area: Responsible AI · Ethics · National governance
Stage: National principles finalised and announced
Date: Finalised 21 August; announced 24 August 2026
Verification: Verified
Source basis: Government primary material
South Korea has finalised national AI Ethics Principles built around human dignity, the public good and sustainability.
The framework also addresses privacy, fairness, accountability, safety, reliability and transparency.
Responsibility is described across the AI ecosystem rather than being assigned exclusively to developers.
Providers, users, government and wider society are all within the framework.
The Principles are expressly voluntary and non-binding.
Their importance lies in their function as a national baseline from which sector-specific guidance and self-assessment tools are expected to develop.
They should not be described as an AI Act, mandatory code or enforcement regime.
South Korea sets out government-wide AI deployment strategy
Jurisdiction: South Korea
Authority: Ministry of the Interior and Safety / Cabinet
Area: Public administration · AI-assisted government · Legal and regulatory review
Stage: Government deployment strategy announced
Date: 25 August 2026
Verification: Verified
Source basis: Government primary publication
South Korea has announced a government-wide AI deployment strategy covering public services and internal administration.
The programme includes 24-hour AI consultations across areas including welfare, taxation, public safety, agriculture and food and drug services, together with personalised public-service assistants and AI-supported disaster analysis.
The government is also developing an internal service known as On-AI.
Of particular interest to the legal record, On-AI is intended to assist civil servants with tasks that include reviewing laws and regulations and drafting reports.
The system is intended to reach 47 central agencies by December 2026, alongside large-scale staff training.
Why it matters
This is operational policy rather than new law.
But the intended use of AI in legal and regulatory review brings it close to the institutional reliance questions raised elsewhere in this edition.
The critical governance question will be whether AI assists officials in locating and analysing material or becomes an unexamined source of propositions entering official decision-making.
The strategy itself does not transfer legal authority from public officials to the system.
United Kingdom and Ukraine establish defence-focused Artificial Intelligence Partnership
Jurisdiction: United Kingdom / Ukraine
Authorities: UK and Ukrainian governments
Instrument: Joint Declaration of Intent
Area: Defence AI · Data · Compute · Assurance · Autonomy · Cybersecurity
Stage: Bilateral political cooperation framework
Date: 24 August 2026
Verification: Verified
Source basis: Government primary declaration
The United Kingdom and Ukraine have signed a Joint Declaration of Intent establishing a defence-focused Artificial Intelligence Partnership.
The contemplated areas of work include co-development of AI models, secure data and compute pathways, joint AI assurance, autonomous systems, cybersecurity, synthetic data and academic and research cooperation.
The declaration adopts a pilot-first approach.
The document is explicit about its legal character: it expresses political intent and does not itself create binding legal obligations.
No particular commercial foundation model, compute supplier or weapons platform is identified.
The development is notable in the same week as Anthropic v U.S. Department of War.
One matter concerns a dispute over the conditions on which a private model provider will permit defence use.
The other concerns governments deliberately building cooperative structures for defence AI, including assurance.
LEGISLATION & RULEMAKING
Mexico's Chamber of Deputies receives proposal covering synthetic media, biometrics and AI-generated works
Jurisdiction: Mexico
Authority: Chamber of Deputies
Area: Criminal law · Copyright · Synthetic media · Biometrics · AI-generated works
Stage: Legislative proposal introduced
Date: 26 August 2026
Verification: Verified at legislative-introduction level
Source basis: Legislative/public-authority record
A legislative initiative introduced in Mexico's Chamber of Deputies proposes amendments to the Federal Criminal Code and Federal Copyright Law addressing a range of AI-related issues.
The proposal covers deceptive synthetic media, specified practices involving biometric and facial data, dissemination of unlawfully obtained material and the treatment of AI-assisted and autonomously AI-generated works under copyright law.
The initiative has entered the legislative process. It is not Mexican law.
No criminal offence, copyright rule or biometric restriction contained in the proposal should be described as operative unless and until enacted through the relevant legislative process.
What happens next
Parliamentary consideration and possible amendment.
OTHER DEVELOPMENTS RECORDED THIS WEEK
The matters below moved during the reporting period or were identified through the backfill.
They are retained publicly because they form part of the developing AI legal and institutional record, but they did not warrant full treatment in this edition.
Courts & litigation
United States — CNN v Perplexity. Perplexity filed a partial motion to dismiss CNN's copyright and trademark action on 21 August. The filing changes no claim by itself; the court has not yet ruled.
United States — Ziff Davis v OpenAI. Ziff Davis filed the conforming amended complaint required by Judge Sidney Stein's 6 August ruling in the OpenAI copyright MDL. The amendment does not revive claims previously dismissed with prejudice.
United States — Alden-associated newspaper litigation against Microsoft and OpenAI. An amended copyright complaint was reportedly filed on 21 August concerning alleged use of newspaper journalism in ChatGPT and Copilot. The exact relationship to the likely canonical California Newspapers Partnership docket remains to be completed against the primary filing.
United States — Barrows v Humana. A joint status report was filed on 24 August in litigation alleging AI-supported influence over Medicare Advantage post-acute-care coverage decisions. The next potentially material event is expected around class-certification proceedings.
United States — Estate of Lokken v UnitedHealth. Scheduling orders were entered on 24 August in litigation concerning alleged use of NaviHealth-related technology in post-acute-care decisions. A hearing on a motion to amend is listed for 1 September.
United States — California ChatGPT product-liability coordination. Co-lead counsel was reportedly appointed on 26 August across a coordinated proceeding comprising 19 ChatGPT product-liability actions. The appointment is procedural and says nothing about defect, causation or liability.
United States — Spirit Aviation bankruptcy. Springshot Inc. filed an objection concerning part of a dataset Spirit proposes to sell to Google for reported uses including AI-model training, arguing that some material may involve Springshot intellectual property. A sale hearing is scheduled for 9 September.
United States — Encyclopaedia Britannica v Perplexity. Discovery-management activity continued concerning retrieval-augmented-generation data, user activity and source code relevant to allegations involving Britannica and Merriam-Webster material.
India — Hyderabad AI-morphed imagery complaint. A criminal complaint was reported after an airline cabin-crew member alleged that sexually explicit manipulated images of her had been created and circulated. The FIR, statutory charges and identity of any accused had not been obtained by the reporting cutoff.
Regulators and public authorities
Australia — AUSTRAC AI Transparency Statement. The financial-intelligence and AML/CTF regulator updated its public account of AI use and governance on 18 August. AUSTRAC says it uses generative AI and AI-enabled analytics while retaining human involvement and has not deployed AI to make administrative decisions without human intervention. The update is governance transparency rather than a new obligation on regulated businesses.
Australia — Department of Defence Cyber Reserve Force. Defence announced a new reserve capability drawing specialists from industry, government and academia, with artificial intelligence identified alongside cyber, cloud, digital forensics and other technical fields. No particular military AI system or deployment was announced.
United Kingdom — MOD Rapid AI Delivery Taskforce. The Ministry of Defence opened an Industry Front Door through which organisations may submit technologies relevant to AI and autonomy. MOD states expressly that the channel is not a procurement exercise, competition or commercial commitment.
England and Wales — College of Policing. A national Microsoft Copilot training programme was launched around the principle “Use. Check. Own.”, emphasising that officers and staff remain responsible for verified outputs and professional judgment.
United Kingdom — Office for National Statistics. ONS announced that it will publish a dedicated AI thematic account of the UK economy. The substantive methodology and figures are expected on 21 September; no AI economic contribution figures were released this week.
United States — Maryland Advisory Committee to the US Commission on Civil Rights. The Committee held an officially noticed meeting on AI in voting administration on 25 August. No substantive post-meeting conclusion had been published at the reporting cutoff.
Australia — National AI Centre. Stakeholder engagement on agentic AI continued during the reporting period. The activity is retained as institutional development rather than treated as a regulatory decision.
South Korea — public procurement. An AI-supported procurement-evaluation pilot continued to develop. Primary documentation remains incomplete; no binding procurement rule is reported from this event.
Nigeria — copyright and AI. Attorney-General Lateef Fagbemi was reported to have called for the Nigerian Copyright Commission to prepare its law, capacity and enforcement approach for technological change including AI and streaming. No formal new rule or primary transcript had been obtained.
Ireland — AI and the labour market. Government published The Irish Labour Market and AI in 2026, examining adoption, employment change and skills demand. It is evidence and policy research rather than new employment regulation.
Uganda — national AI strategy. The ICT Minister reportedly told Parliament that a draft National AI and Emerging Technologies Strategy is under ministry review before possible Cabinet consideration. It has not been adopted.
Sri Lanka — legislative drafting. A government workshop examined possible use of AI-assisted drafting and validation tools within the Legal Draftsman's Department, including a possible pilot involving the Virtual Asset Service Providers Bill. No AI system has been delegated legislative authority.
Namibia — agricultural monitoring contract. Government terminated an approximately N$39.5 million AI- and satellite-based agricultural monitoring agreement with 6th Grain Corporation. Available accounts attribute the decision to concerns around the contracting process rather than findings about AI performance.
International and intergovernmental activity
BRICS — Jaipur Declaration. BRICS Tourism Ministers adopted the Jaipur Declaration on 21 August, including responsible, safe and human-centred AI as a cooperation priority in tourism. The declaration is political and intergovernmental rather than directly enforceable AI law.
India — healthcare AI. Union Minister of State Anupriya Patel used the IIMA Healthcare Summit to state that AI should support areas including drug discovery, screening, surveillance and clinical decision-making while remaining safe, privacy-protective and complementary to human clinical judgement. No new binding healthcare-AI instrument was issued.
United Arab Emirates — Abu Dhabi Department of Energy. The Department announced AI-related cooperation memoranda and explored creation of an AI-powered laboratory for energy and water technologies. The agreements are development arrangements; the proposed laboratory has not yet been established.
International — BIS/Irving Fisher Committee. The 13th Biennial Conference on central-bank statistics included substantive work on generative AI, machine learning and AI-supported financial and economic analysis. No new BIS supervisory standard emerged from the conference.
Lesotho — national AI strategy development. A government and United Nations-supported workshop concluded as part of development of a National AI Strategy and Data Centre Blueprint. Completion of the workshop does not mean either instrument has been formally adopted.
DEVELOPMENTS TO WATCH
The Watch list is deliberately selective. It identifies matters with a meaningful forthcoming trigger, deadline or unresolved event rather than reproducing every technically live matter in the underlying database.
Imminent
Project Jupiter — New Mexico Supreme Court
Interim air-quality and water stays entered. Responses reportedly due 2 September 2026.
Estate of Lokken v UnitedHealth
Hearing on motion to amend listed for 1 September 2026.
Spirit Aviation bankruptcy / Springshot objection
Sale hearing listed for 9 September 2026.
Australian Joint Select Committee on Artificial Intelligence
Submissions close 14 September 2026.
Northern Ireland — Digital Omnibus on AI
Committee report published; wider scrutiny period currently runs to 24 September 2026.
Autumn and beyond
ONS AI thematic account
Substantive publication expected 21 September 2026.
Singapore AI and IP consultation
Submissions close 22 October 2026 at 5:00pm Singapore time.
Australian Joint Select Committee on Artificial Intelligence
Report due 30 November 2026.
Australia — national AI/data-centre framework
Commonwealth legislation currently intended for early 2027.
Awaiting a trigger
Anthropic v U.S. Department of War
Government compliance, stay application or appeal.
Taiwan AI-server prosecutions
Criminal-court allocation and first hearing.
Faiz Enterprise
Full Gujarat High Court judgment and any fresh tax process.
Justice v Suno
Signed 20 August opinion and next responsive-pleading deadline.
Alabama/OpenAI — ALR-0047
OpenAI's response to compulsory investigative process and any subsequent enforcement decision.
X.AI v Ellison — ALR-0009
Preliminary-injunction and dismissal rulings.
Amazon v Perplexity
Ninth Circuit decision on rehearing en banc.
Dutch DPA / Twitch-Amazon
Any formal investigation, platform change or enforcement response.
South Korea FSS
Primary regulator documentation and further on-chain or fund-flow surveillance capability.
Alden newspaper litigation
Primary amended complaint and definitive canonical docket reconciliation.
Reported Hyderabad AI-morphed imagery matter
FIR, charges and any accused to be confirmed.
Reported new Jane Doe/xAI litigation concerning Grok
No distinct complaint and docket had been obtained by the reporting cutoff. The matter is not treated as a verified lawsuit until primary documentation is located.
Dubois v Traxsource / Thrust Publishing
Reported dispute concerning music allegedly misclassified as AI-assisted remains pending retrieval of the amended complaint.
In re Mosaic LLM Litigation
Reported amicus application concerning fair use and LLM training remains subject to docket confirmation.
METHODOLOGY & VERIFICATION NOTE
The AI Law Report separates discovery, verification, reconciliation, database enrichment and publication.
Developments are identified through daily global monitoring across two principal streams: courts and litigation, and regulators and public authorities.
Discovery runs on a rolling short-term window with a seven-day backfill so that filings, judgments and authority publications that surface late are not lost merely because they were indexed after the day on which they occurred.
Candidates are then reconciled against the existing Law Report record.
That distinction matters increasingly as the database grows.
A complaint, hearing, amended pleading, order, appeal and judgment can all be separate events within the same legal matter.
They do not become separate matters merely because they occur in different weeks.
Permanent ALR references are therefore intended to identify the underlying canonical matter rather than every individual event within it.
Matter numbers are assigned once, in the order a matter enters the Report's underlying record, and are never reused or reassigned.
A number may therefore be assigned at the time of the underlying event or, where an older matter is newly entered into the record, at the point of backfill.
Issues 002 and 003 introduced the public ALR numbering system.
Issue 001 preceded it.
The master register is currently being reconciled across Issues 001–004. Existing confirmed ALR references are retained in this edition. New matter references will be added after that reconciliation without changing the substantive reporting.
A matter may remain live within the internal database without appearing in every subsequent published Watch list.
The publication Watch list is reserved for matters with a meaningful forthcoming deadline, unresolved material event or heightened editorial significance.
No matter is considered closed or dormant internally merely because it is absent from the weekly publication.
Publication threshold
Discovery is deliberately broad.
Publication is not.
A development receives full treatment where it materially advances a judicial proceeding, creates or clarifies a legal rule, articulates a meaningful supervisory expectation, begins or escalates substantive enforcement, or establishes a public-authority position sufficiently concrete to affect how AI is governed, deployed or relied upon.
This edition introduces Other Developments Recorded This Week for matters that pass the relevance threshold but not the threshold for full editorial treatment.
The purpose is not to turn the Report into a chronology of every AI conference, training course, workshop or procurement notice.
It is to preserve publicly the movement of legally relevant matters without giving routine procedure the same editorial weight as a judgment, enforcement action or substantive regulatory development.
Verification
Verification is reported rather than assumed.
Where a primary judgment, filing, legislative text, regulatory publication or other authoritative source has been obtained, that forms the preferred basis of the entry.
Where the underlying development is strongly corroborated but the primary document is unavailable, the limitation is stated and the prose is correspondingly narrower.
A significant development is not automatically omitted because a signed judgment has not yet reached a searchable public database.
But reporting does not convert secondary accounts into judicial findings.
Two source limitations carry particular weight this week.
The complete signed merits judgment in Anthropic v U.S. Department of War has not yet been retained.
The result and principal grounds are strongly established, but the exact operative wording of the injunction should remain tied to the final order.
Likewise, the full Gujarat High Court judgment in Faiz Enterprise has not yet been obtained.
The reported admission of AI use and the quashing of the tax process are sufficiently corroborated for publication, but broader reported directions concerning future AI use should not be stated more precisely until the judgment is available.
Matters deliberately not treated as AI-law developments
The inclusion test also operates in the other direction.
The Dutch Data Protection Authority's approximately €825 million Uber decision concerning solely automated driver-account decisions is a major data-protection development.
However, the material reviewed establishes automated or algorithmic decision-making and does not establish that the relevant decision system used artificial intelligence or machine learning.
Automation is not automatically AI.
The matter is therefore retained outside the accepted AI register unless that nexus is later established.
The US Federal Trade Commission's proposed enforcement policy on personalised pricing is likewise a genuine regulatory development.
But the proposal applies to personalised pricing based on personal data whether or not AI is used.
The FTC's wider surveillance-pricing work discusses AI and machine learning; that does not make AI a necessary element of this particular policy.
Finally, reporting on the Australian Fair Work Commission's use of agentic AI during this period was not treated as a new deployment event.
Verification showed that the underlying systems had already been publicly documented in July.
A fresh article is not a fresh legal event.
Editorial rules
Allegations remain allegations.
A complaint is not a finding.
A motion is not a ruling.
A referral is not a disciplinary determination.
An indictment is not a conviction.
A consultation is not regulation.
A legislative proposal is not law.
A political declaration is not a binding obligation.
An interim stay is not a final merits judgment.
A system being used by a regulator does not mean the system is making the regulator's decisions.
And the existence of automation does not, without evidence, establish the use of artificial intelligence.
Corrections are published visibly rather than silently incorporated into the historical record.
Coverage during this reporting period materially searched the United States and Canada; the United Kingdom and Ireland; EU institutions and national European jurisdictions; India and South Asia; East and Southeast Asia; Australia and New Zealand; the Middle East and Africa; and Latin America and the Caribbean.
Near-real-time access to lower-court dockets and official publications remains uneven across jurisdictions.
The absence of a development from a country in a particular edition should not be read as evidence that none occurred.
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, legislation, regulatory publications and other primary sources.
Where primary documentation could not be located, this is stated in the relevant entry.
Editorial update on ALR references: Issue 004 is being published while the Report's master matter register is undergoing reconciliation across Issues 001–004. Existing confirmed ALR references are retained. New matter references will be added to this edition once that reconciliation is complete. The addition of those references will not alter the substantive reporting.
Next edition: Friday 4 September 2026.