The AI Law Report Issue 006 - Friday 11 September
Issue 006 | Friday 11 September 2026
Published by AI Policies UK
This edition covers developments identified between 5 and 11 September 2026. Earlier decisions and publications followed up during the week appear separately, with their original event dates. Sources were checked on 11 September 2026.
This week
California enacted laws on companion-chatbot child safety, independent AI verification and AI audit services. China’s Supreme People’s Court issued guidance on AI disputes, and Canada’s Immigration and Refugee Board brought its AI practice notice into force. In the United Kingdom, Ofcom opened an enforcement programme on intimate-image abuse and the healthcare AI commission published recommendations for regulatory reform.
Court and professional-conduct coverage includes an Ohio sentence involving synthetic sexual imagery, a Pennsylvania cyberstalking indictment, copyright proceedings and a New Brunswick lawyer’s suspension following fictitious AI-generated citations. The report also records Australian draft legislation, US congressional oversight and developments involving public authorities in Singapore, Canada, China and the European Union.
Each development has a permanent publication reference in the form ALR 2026/006/01. This is separate from the court, legislative or authority reference recorded below it. Earlier editions will be assigned corresponding ALR references in the quarterly consolidated index.
Courts and litigation
James Strahler II sentenced to 15 years in case involving AI sexual imagery
ALR reference: ALR 2026/006/01 | Status: VERIFIED — NEW MATTER
Jurisdiction: United States — federal, Ohio
Court / authority: US District Court for the Southern District of Ohio (Chief Judge Sarah D. Morrison)
Case / instrument: United States v James Strahler II
Case number / official reference: 2:26-cr-00038
Area: Federal criminal law · Cyberstalking · Digital forgeries · Child sexual abuse material
Stage: Guilty pleas entered in April 2026; sentence imposed
Event date: 8 September 2026 | Confidence: High
What happened. Chief US District Judge Sarah D. Morrison sentenced James Strahler II to 180 months’ imprisonment after his April guilty pleas to cyberstalking, producing obscene visual representations of child sexual abuse, and publication of digital forgeries. The Department of Justice described the case as its first prosecution and conviction under the TAKE IT DOWN Act, enacted in 2025.
According to the department’s account, Strahler’s conduct between December 2024 and June 2025 included harassing at least six adult women through messages, calls and online postings. He circulated real and AI-generated intimate material, threatened victims and contacted their relatives. Prosecutors also described synthetic sexual material made using the faces of children from his community. Police referred the investigation to the FBI, and he was arrested on federal charges in June 2025.
The digital-forgery conviction forms part of the TAKE IT DOWN Act prosecution. The 15-year sentence covers the combined offences, rather than that offence alone. The account comes from the department’s sentencing announcement.
The underlying question. Sentencing following admitted federal offences involving real and AI-generated intimate imagery.
What this development does not establish. The 15-year term covers the combined offences before the court. It is not a sentence solely for the TAKE IT DOWN Act count.
What happens next. No appeal or other post-sentence filing had been identified at the editorial cut-off.
Verification. Verified against the US Attorney’s Office sentencing announcement. The formal sentencing judgment was not separately reviewed.
Source: US Department of Justice sentencing announcement, 8 September
Scott Testa indicted over alleged cyberstalking with AI-altered images
ALR reference: ALR 2026/006/02 | Status: VERIFIED — NEW MATTER
Jurisdiction: United States — federal, Pennsylvania
Court / authority: US District Court for the Eastern District of Pennsylvania; US Magistrate Judge Scott W. Reid
Case / instrument: United States v Scott Testa
Case number / official reference: No public case number identified in the source reviewed
Area: Federal criminal law · Cyberstalking · AI-altered imagery
Stage: Indictment, arrest and initial appearance; defendant detained pending trial
Event date: 8 September 2026 | Confidence: High as to the announced procedural events
What happened. Federal prosecutors announced that Scott Testa, 60, of Lafayette Hill, had been arrested and indicted on two counts of cyberstalking. The charges concern alleged online conduct targeting a local government official, a police officer and their families in Whitpain Township.
The indictment, as described by prosecutors, alleges that April 2026 Reddit posts used AI-altered images to depict the victims and family members in Ku Klux Klan robes. Posts allegedly included identifying details, the names of children, an address believed to belong to a victim, false information and threatening messages. Prosecutors allege a course of conduct intended to harass and intimidate, causing or attempting to cause substantial emotional distress.
US Magistrate Judge Scott W. Reid ordered Testa detained in federal custody pending trial at a hearing on 8 September. The detention decision does not determine the charges. The published account is a prosecution announcement of allegations in an indictment; Testa is presumed innocent unless and until proved guilty.
The underlying question. Whether the alleged online course of conduct satisfies the federal cyberstalking offences charged in the indictment.
What this development does not establish. The allegations have not been proved. The detention order is not a determination of guilt.
What happens next. Pre-trial proceedings will follow. No trial date was identified in the material reviewed.
Verification. Verified against the US Attorney’s Office announcement. The indictment PDF and its case number were not separately retrieved.
Source: US Department of Justice charging announcement, 8 September
Association of Corporate Counsel brings copyright and trademark claims against The L Suite
ALR reference: ALR 2026/006/03 | Status: PARTIALLY VERIFIED — NEW MATTER
Jurisdiction: United States — federal, Delaware
Court / authority: US District Court for the District of Delaware
Case / instrument: Association of Corporate Counsel v The L Suite
Case number / official reference: 1:26-cv-01138, as reported
Area: Copyright · Trademark · Generative legal AI
Stage: Civil complaint filed; no liability determination
Event date: 10 September 2026, as reported | Confidence: Medium-high
What happened. Reuters reports that the Association of Corporate Counsel sued The L Suite over alleged use of ACC materials to train its Lloyd chatbot and in marketing. The reported action, case 1:26-cv-01138, includes copyright and trademark claims and seeks damages and injunctive relief.
The claims concern commercial use of a professional association’s material. No infringement finding is established. This account relies on Reuters’ description of the complaint.
The underlying question. Whether the alleged use of ACC material in training, developing or promoting the Lloyd chatbot infringes ACC’s copyright or trademark rights.
What this development does not establish. The filing does not establish infringement, misuse of a mark or entitlement to the remedies sought.
What happens next. Service, responsive pleadings and case management. No response deadline was identified in the source reviewed.
Verification. The action and reported case number are supported by Reuters. The complaint and official docket were not directly inspected.
Source: Reuters report, 10 September
Oklahoma Attorney General declines criminal prosecution of Judge Lawrence Wheeler
ALR reference: ALR 2026/006/04 | Status: PARTIALLY VERIFIED — PROSECUTORIAL DECISION
Jurisdiction: United States — Oklahoma
Court / authority: Oklahoma Attorney General; underlying matter concerns District Judge Lawrence Wheeler
Case / instrument: Criminal review concerning Judge Lawrence Wheeler
Case number / official reference: No public reference identified
Area: Judicial conduct · Criminal review · Generative-AI legal research
Stage: Attorney General declined criminal prosecution; separate disciplinary position unresolved
Event date: 9 September 2026 | Confidence: Medium-high
What happened. Reuters reports that Oklahoma’s Attorney General said the evidence concerning Judge Lawrence Wheeler was insufficient for criminal prosecution. The report describes an order containing two fictitious citations generated using ChatGPT, alongside separate allegations concerning the judge’s conduct.
The decision reported on 9 September concerns whether to bring criminal proceedings. It is not a finding that the citations were authentic, and it does not establish that any professional-disciplinary process has cleared the judge. This entry records the prosecutorial decision as reported by Reuters.
The underlying question. Whether the evidence referred to the Attorney General justified criminal proceedings in a wider judicial-conduct matter that included alleged ChatGPT-generated authorities.
What this development does not establish. The decision is not an acquittal, a disciplinary exoneration or a finding that the use of ChatGPT in the judicial order was professionally acceptable.
What happens next. Any separate action by Oklahoma’s judicial-conduct authorities. No future hearing or disciplinary date was identified.
Verification. The no-prosecution decision is supported by Reuters reporting based on confirmation from the Attorney General’s Office. The underlying order and investigation materials were not obtained.
Source: Reuters report, 9 September
Court procedure and judicial guidance
Canada’s immigration tribunal brings AI practice notice into force
ALR reference: ALR 2026/006/05 | Status: VERIFIED — PROCEDURAL INSTRUMENT
Jurisdiction: Canada — federal administrative tribunal
Court / authority: Immigration and Refugee Board of Canada, all four divisions
Case / instrument: Practice notice: Use of artificial intelligence in IRB proceedings
Case number / official reference: No separate instrument number identified
Area: Tribunal procedure · Evidence · Professional responsibility · Disclosure
Stage: Practice notice in force
Event date: 7 September 2026; signed 10 July 2026 | Confidence: High
What happened. The Immigration and Refugee Board’s AI practice notice took effect across all four divisions on 7 September. Parties and counsel remain responsible for submitted material, including self-represented parties.
The notice prohibits AI generation or material alteration of personal evidence, including claim narratives, affidavits, witness statements and corroborative photographs or videos. Minor spelling, grammar and formatting assistance remains permitted if it does not generate evidence or materially change it. Legal principles and citations must be checked against reliable sources; AI summaries are insufficient verification.
Documents containing AI-generated or materially changed wording require a declaration identifying the person who verified the content and authorities. AI translation and transcription must always be disclosed. Translated documents still require a human translator’s declaration.
Non-compliance may lead to rejection of a document, adverse credibility inferences, disclosure to counsel’s professional regulator, or restrictions on counsel appearing before the board. Parties must also explain their AI use when asked.
The underlying question. How parties and counsel may use AI when preparing evidence, submissions, translations and legal research for IRB proceedings.
What this development does not establish. The notice does not prohibit minor spelling, grammar or formatting assistance that neither generates evidence nor materially changes it.
What happens next. The requirements now apply in IRB proceedings. No scheduled review date was announced.
Verification. Verified from the complete practice notice published by the IRB.
Source: IRB Practice Notice on the Use of Artificial Intelligence in Proceedings
China’s Supreme People’s Court issues 24 provisions on AI disputes
ALR reference: ALR 2026/006/06 | Status: VERIFIED — JUDICIAL GUIDANCE
Jurisdiction: China — national
Court / authority: Supreme People’s Court of the People’s Republic of China
Case / instrument: Opinions on Lawfully Adjudicating Disputes Involving Artificial Intelligence
Case number / official reference: Fa Fa [2026] No. 10
Area: AI disputes · Personality rights · Data · Product liability · Copyright · Evidence
Stage: Twenty-four judicial provisions issued
Event date: 7 September 2026 | Confidence: High
What happened. The Supreme People’s Court issued Opinions on Lawfully Adjudicating Disputes Involving Artificial Intelligence, Fa Fa [2026] No. 10. Its 24 provisions address the application of existing law to disputes involving AI products, services and outputs.
The guidance addresses responsibility for misuse of likenesses and voices, personal-information processing, defective products and copyright infringement. Liability assessments consider matters including autonomy, control, risk and preventive measures. Public availability of personal information does not remove all restrictions on its subsequent use.
For copyright disputes, courts are directed to consider participants’ roles, sources of data, benefits and measures taken against infringement. A developer relying on a training-related defence may need to produce evidence about training data, processes and model operation. The text also addresses providers’ responses to infringement notices and users’ deliberate use of infringing prompts.
Procedural provisions cover evidence production and verification of AI-assisted court submissions, including fabricated authorities or evidence. These directions apply existing law; they neither decide a case nor enact an AI statute.
The underlying question. How courts should apply existing legal rules to disputes involving AI products, services, training material, outputs and evidence.
What this development does not establish. The Opinions do not determine liability in an individual case and do not enact a comprehensive AI statute.
What happens next. The guidance is available for application in relevant proceedings. No separate future commencement date was identified.
Verification. Verified from the Supreme People’s Court publication and full Chinese text.
Source: Supreme People’s Court Opinions, 7 September, Chinese text
Regulators and enforcement
Ofcom opens enforcement programme on illegal intimate images and deepfakes
ALR reference: ALR 2026/006/07 | Status: VERIFIED — ENFORCEMENT PROGRAMME
Jurisdiction: United Kingdom
Court / authority: Ofcom
Case / instrument: Enforcement programme concerning non-consensual intimate images and explicit AI deepfakes
Case number / official reference: No separate enforcement reference identified
Area: Online safety · Non-consensual intimate imagery · Deepfakes · Platform duties
Stage: Enforcement programme opened; relevant code measures approaching application
Event date: 9 September 2026 | Confidence: High
What happened. Ofcom opened a programme to monitor platforms’ compliance with duties addressing non-consensual intimate images, including explicit AI-generated deepfakes. Its announcement identifies 30 September as the deadline for relevant measures to prevent this material spreading.
Platforms should use hash matching or demonstrate equally effective alternative systems. Ofcom announced cooperation with SWGfL, which operates StopNCII.org and the Revenge Porn Helpline, to share evidence and expertise. It said failure to comply with legal duties can attract fines of up to 10% of global annual revenue.
Separately, Ofcom plans to consult by the end of 2026 on amendments to its Illegal Harms Codes reflecting the law on removal within 48 hours of a report.
The regulator also invited evidence on services’ response to harms affecting women and girls, including harassment and privacy risks. Responses are due by 6 November, with a progress report planned for summer 2027.
The underlying question. Whether regulated services have effective systems to prevent the spread of illegal intimate images, including explicit AI-generated deepfakes.
What this development does not establish. Opening the programme is not a finding that an individual service has breached the Online Safety Act. Hash matching is not the only permitted method where an equally effective alternative can be demonstrated.
What happens next. Relevant measures are identified for 30 September. Ofcom plans a further codes consultation by the end of 2026; its evidence invitation closes on 6 November.
Verification. Verified from Ofcom’s enforcement announcement.
Source: Ofcom enforcement announcement, 9 September
San Francisco demands action from Meta over AI abuse advertisements
ALR reference: ALR 2026/006/08 | Status: PARTIALLY VERIFIED — ENFORCEMENT DEMAND
Jurisdiction: United States — California, San Francisco
Court / authority: Office of the San Francisco City Attorney
Case / instrument: Demand to Meta concerning advertisements for AI abuse tools
Case number / official reference: Underlying demand-letter reference not identified
Area: Platform responsibility · Advertising · Child sexual abuse material · AI image tools
Stage: Demand reported; no court order identified
Event date: 9 September 2026, as reported | Confidence: Medium
What happened. WIRED reports that San Francisco’s City Attorney demanded that Meta stop allowing advertisements for tools associated with AI-generated child-sexual-abuse material. The report describes a dispute over the advertisements and Meta’s responsibility for carrying them.
The development is a reported enforcement demand. It is not a court order or an adjudicated finding that Meta has committed the alleged violations. The underlying demand letter was not inspected for this edition.
The underlying question. Whether Meta’s carrying of advertisements for the identified tools engages the legal and platform-responsibility concerns asserted by the City Attorney.
What this development does not establish. The reported demand is not a court order or an adjudicated finding that Meta committed a violation.
What happens next. Meta’s response and any formal enforcement proceeding. No deadline or filed case was verified.
Verification. Supported by WIRED’s report. The City Attorney’s underlying demand letter was not directly inspected.
Source: WIRED report, 9 September
SEC publishes AI and disclosure materials for Investor Advisory Committee meeting
ALR reference: ALR 2026/006/09 | Status: VERIFIED — PUBLIC-AUTHORITY MATERIALS
Jurisdiction: United States — federal
Court / authority: Securities and Exchange Commission Investor Advisory Committee
Case / instrument: AI and the information available to public-market investors: agenda and presentations
Case number / official reference: IAC meeting of 10 September 2026; no rulemaking number
Area: Securities disclosure · Structured data · Investor information · AI-assisted analysis
Stage: Agenda and presentation materials published
Event date: 10 September 2026 | Confidence: High as to the published materials
What happened. The SEC published presentation materials for its Investor Advisory Committee meeting of 10 September. The agenda included a panel on AI and the information available to public-market investors, covering disclosure, access, data quality and the risk of relying on automated outputs.
A published presentation by Julie Marlowe of the SEC’s Office of Structured Disclosure addresses machine-readable disclosures, structured data and the information needed to interpret company reporting. Her slides address how standardised fields and semantic context can support AI use, alongside complications arising from inconsistent or customised tagging and changes to reporting requirements.
A separate presentation by Campbell Pryde of XBRL US examines structured company data in AI-assisted financial analysis. His material argues that defined data and contextual relationships can help distinguish financial concepts that are difficult to recover reliably from unstructured documents.
The published documents establish the content of the presentations. They are not a Commission rule, an adopted committee recommendation or a transcript establishing the full discussion. The views expressed in individual presentations remain attributable to their authors.
The underlying question. How disclosure structure, semantic context, data quality and access affect AI-assisted analysis of public companies.
What this development does not establish. The materials are not an SEC rule, an adopted committee recommendation or a transcript of the complete discussion. Presenters’ views remain attributable to them.
What happens next. Any subsequent committee recommendation or Commission action. None was announced in the reviewed materials.
Verification. Verified from the SEC agenda and the published presentations by Julie Marlowe and Campbell Pryde.
Sources: SEC Investor Advisory Committee: agenda and materials, 10 September | Julie Marlowe: SEC disclosure and AI presentation | Campbell Pryde: XBRL US presentation
ESMA includes frontier-AI risks in financial-market monitoring
ALR reference: ALR 2026/006/10 | Status: VERIFIED — RISK-MONITORING REPORT
Jurisdiction: European Union
Court / authority: European Securities and Markets Authority
Case / instrument: Trends, Risks and Vulnerabilities Report, No. 2, 2026
Case number / official reference: ESMA50-1949966494-4282
Area: Financial markets · Operational resilience · Cyber risk · Frontier AI
Stage: Risk-monitoring report published
Event date: 10 September 2026 | Confidence: High
What happened. ESMA published its second risk-monitoring report of 2026, covering vulnerabilities in EU financial markets. The authority said strong performance in technology and AI-related sectors had supported investor optimism despite weaker macroeconomic conditions and heightened geopolitical uncertainty.
The release identifies emerging frontier-AI threats to market infrastructure and other participants. It describes cyber risk as an increasingly important part of the operational-risk environment, while also recording continued growth in AI-focused investment funds, particularly funds targeting infrastructure.
These are ESMA’s assessments of market conditions and exposures. The publication does not itself impose new AI-specific compliance duties or announce a finding that an individual firm has breached securities law.
The underlying question. How frontier-AI developments, cyber threats and AI-related valuations feature in ESMA’s assessment of EU financial-market vulnerabilities.
What this development does not establish. The risk assessment does not itself impose a new AI-specific compliance duty or establish a breach by an individual firm.
What happens next. ESMA’s monitoring continues. No AI-specific rulemaking deadline was announced in this publication.
Verification. Verified from ESMA’s official release and report listing.
Source: ESMA risk-monitoring release, 10 September
Legislation and proposals
California signs Adam’s Law on companion-chatbot child safety
ALR reference: ALR 2026/006/11 | Status: VERIFIED — ENACTED LEGISLATION
Jurisdiction: United States — California
Court / authority: California Legislature and Governor
Case / instrument: SB 1119, Adam’s Law: companion chatbots and children’s safety
Case number / official reference: Senate Bill No. 1119
Area: Child safety · Companion chatbots · Consumer protection · Audits
Stage: Signed; principal child-safety provisions not yet operative
Event date: 10 September 2026 | Confidence: High
What happened. Governor Gavin Newsom signed SB 1119, known as Adam’s Law, as part of California’s child-safety package. The Governor’s announcement identifies parental controls, crisis-response protocols, risk assessments and independent child-safety audits among its measures for companion chatbots.
The enacted text establishes safeguards for operators allowing children to use covered services, including restrictions on sexual content and relationship-manipulating responses. Statutory defaults limit use to one continuous hour and two hours daily, subject to parental-control provisions. Crisis protocols address credible imminent risks and include conditions governing parental notification and access to crisis support.
The principal child-safety provisions become operative on 1 July 2027. The first independent audit is required by 1 January 2029 or before the chatbot is first made available, whichever is later, followed by biennial audits and specified reassessments. Operators below US$500 million in annual gross revenue have an audit exemption until 1 January 2032.
The law provides public enforcement and a private right of action for actual harm caused by specified violations. That private remedy does not extend automatically to every requirement in the chapter. Signing the legislation does not bring all of its duties into operation immediately.
The underlying question. What duties covered companion-chatbot operators owe where children use their services, including controls, crisis protocols, assessments and audits.
What this development does not establish. Signing does not bring every duty into immediate operation. The private right of action is tied to actual harm from specified violations and does not extend automatically to the whole chapter.
What happens next. Principal child-safety provisions become operative on 1 July 2027. The conditional first-audit deadline is 1 January 2029 or first availability, whichever is later.
Verification. Verified against the enrolled legislative text and the Governor’s signing announcement.
Sources: California Governor’s signing announcement, 10 September | California SB1119 legislative text
California creates a framework for independent AI verification organisations
ALR reference: ALR 2026/006/12 | Status: VERIFIED — ENACTED LEGISLATION
Jurisdiction: United States — California
Court / authority: California Government Operations Agency; California Legislature and Governor
Case / instrument: SB 813, Independent verification organizations
Case number / official reference: Senate Bill No. 813; Chapter 179
Area: AI assurance · Independent verification · Audit standards · Evidence
Stage: Signed; agency framework to be developed
Event date: 9 September 2026 | Confidence: High
What happened. SB 813 requires California’s Government Operations Agency to establish requirements and criteria for designated independent verification organisations by 1 January 2028. Designation concerns expertise in assessing AI-system or model risks and identifying the metrics and methods used for that assessment.
The agency must consider technical competence, independence and conflicts of interest. An organisation may charge reasonable market rates, but payment cannot depend on the outcome of its assessment. The framework includes procedures for suspension or termination of designation, consultation through working groups and annual reporting by designated organisations.
The statute expressly states that it does not require every developer, deployer or operator to engage a verification organisation or undergo an audit. Publication of criteria does not amount to state endorsement of a system or model.
Where harm from AI is alleged, an audit conducted against an identified standard is relevant evidence but is not conclusive.
The underlying question. What criteria California should apply when designating organisations to assess AI-system risks and the methods used for those assessments.
What this development does not establish. The statute does not require every AI developer or deployer to use an IVO, and designation does not amount to state endorsement of an AI system or model.
What happens next. The Government Operations Agency must undertake the specified framework actions by 1 January 2028 and convene stakeholder working groups.
Verification. Verified against California’s chaptered legislative text.
Source: California SB813 legislative text
California establishes AI auditor registration and reporting requirements
ALR reference: ALR 2026/006/13 | Status: VERIFIED — ENACTED LEGISLATION
Jurisdiction: United States — California
Court / authority: California Government Operations Agency; California Legislature and Governor
Case / instrument: AB 1405, Artificial intelligence: auditors: registration
Case number / official reference: Assembly Bill No. 1405; Chapter 178
Area: AI audits · Registration · Independence · Professional standards
Stage: Signed; registry and service restrictions scheduled
Event date: 9 September 2026 | Confidence: High
What happened. AB 1405 requires the Government Operations Agency to establish an online AI Auditor Registry by 1 January 2029. From that date, a person must be registered to offer, sell or conduct a covered audit: an assessment of AI-related controls, processes or systems necessary for compliance with state law.
Registered auditors must disclose information about their services and methods. Reports must state their scope, results and supporting basis, identified deficiencies, limitations and material evidence or access gaps. The legislation requires a signed compliance statement and retention of relevant records for at least ten years.
Independence provisions address financial and employment relationships, auditing one’s own work and conflicts affecting professional judgement. Whistleblower protections prohibit specified restrictions and retaliation. The agency may investigate alleged violations, remove an auditor from the register and refer matters for enforcement, subject to procedural safeguards.
Specified accountancy professionals can satisfy certain requirements through applicable professional standards. Registration does not constitute state endorsement. The Act regulates covered audit services; its definition does not encompass every informal evaluation of AI.
The underlying question. Who may offer covered AI audits for compliance with California law and what independence, reporting, record-retention and integrity requirements apply.
What this development does not establish. Registration does not constitute state endorsement. The statutory definition of a covered audit does not encompass every informal assessment of AI.
What happens next. The registry must be established no later than 1 January 2029. From that date, covered audit services require registration, subject to the Act’s provisions.
Verification. Verified against California’s chaptered legislative text.
Source: California AB1405 legislative text
California enacts separate measures on chatbot toys and synthetic child-abuse material
ALR reference: ALR 2026/006/14 | Status: VERIFIED — ENACTED LEGISLATION
Jurisdiction: United States — California
Court / authority: California Legislature and Governor
Case / instrument: SB 867 and SB 1276: chatbot toys and synthetic child-abuse material
Case number / official reference: Senate Bill Nos. 867 and 1276
Area: Toy safety · Companion chatbots · Child sexual exploitation · Synthetic media
Stage: Both measures signed
Event date: 10 September 2026 | Confidence: High
What happened. Two further enactments in the child-safety package address distinct products and conduct. SB 867 establishes a prohibition concerning the manufacture, sale, offering for sale or distribution of toys with companion-chatbot functionality, lasting until 1 January 2031. The measure is directed at toys within its statutory scope, rather than every conversational AI service.
SB 1276 amends provisions on the sexual exploitation of children and associated reporting. Its changes include digitally altered or AI-generated matter depicting a person under 18 engaged in specified sexual conduct. Covered conduct extends to downloading, streaming and access through electronic or digital media, subject to the legislation’s knowledge requirements and exceptions.
The two measures have separate statutory definitions and provisions from SB 1119’s companion-chatbot operating safeguards.
The underlying question. How California addresses companion-chatbot functionality in toys and digitally altered or AI-generated sexual material depicting children.
What this development does not establish. The two statutes have different definitions and legal effects. They do not form a single regime with SB 1119 merely because they were announced in the same package.
What happens next. Implementation follows each statute’s operative provisions. No additional common administrative deadline was identified.
Verification. Verified against both legislative texts and the Governor’s signing announcement.
Sources: California SB867 legislative text | California SB1276 legislative text | California Governor’s signing announcement, 10 September
California restricts the use of covered student information in AI development
ALR reference: ALR 2026/006/15 | Status: VERIFIED — ENACTED LEGISLATION
Jurisdiction: United States — California
Court / authority: California Legislature and Governor
Case / instrument: AB 1159, Student personal information
Case number / official reference: Assembly Bill No. 1159
Area: Education · Student information · Generative-AI training · Privacy
Stage: Signed; higher-education regime has a future operative date
Event date: 10 September 2026 | Confidence: High
What happened. AB 1159 adds restrictions on educational-service operators’ use of covered student information to train generative AI or develop AI systems. It also extends aspects of the operator framework to third parties.
The restrictions apply within the statutory definitions of operators, educational services and covered information.
AB 1159 also establishes a higher-education information-protection regime with an operative date of 1 July 2027. It was signed as part of the 10 September child-safety package.
The underlying question. When covered educational-service operators and third parties may use specified student information for generative-AI training or AI development.
What this development does not establish. The measure is defined by its covered operators, educational services and information. It is not a general prohibition on every use of AI in education.
What happens next. The higher-education information-protection regime becomes operative on 1 July 2027.
Verification. Verified against the legislative text and the Governor’s signing announcement.
Sources: California AB1159 legislative text | California Governor’s signing announcement, 10 September
Australia releases draft Digital Duty of Care legislation
ALR reference: ALR 2026/006/16 | Status: VERIFIED — EXPOSURE DRAFT
Jurisdiction: Australia — federal
Court / authority: Australian Government
Case / instrument: Digital Duty of Care exposure draft
Case number / official reference: No parliamentary bill number assigned
Area: Online safety · Children · AI chatbots · Algorithmic feeds · Nudify services
Stage: Exposure draft released for targeted consultation
Event date: 8 September 2026 | Confidence: High
What happened. The Australian Government released draft Digital Duty of Care legislation for targeted consultation. The proposals include duties for AI chatbots, games and apps to protect people under 18 from harmful content and design features. The government says platforms would need to document, implement and maintain measures addressing the relevant risks.
The package also proposes giving social-media users a choice over algorithmically personalised default feeds. The eSafety Commissioner would receive powers to issue removal notices concerning nudify apps or websites. The government’s announcement identifies potential penalties of up to A$109.2 million for non-compliance.
The publication is an exposure draft. The government intends to introduce legislation to Parliament later in 2026; the duties, powers and penalties announced remain proposals.
The underlying question. What preventative duties online services, including AI chatbots, games and apps, should owe to users under 18, and what additional platform powers should be available.
What this development does not establish. The announced duties, removal powers and penalties remain proposals. They have not passed Parliament and are not currently enforceable under this draft.
What happens next. The Government intends to introduce legislation later in 2026. The reviewed announcement did not supply a final enactment timetable.
Verification. Verified from the Australian Prime Minister’s official announcement.
Source: Australian Prime Minister’s draft-law announcement, 8 September
Government and public authorities
UK healthcare AI commission recommends staged authorisation and continuing oversight
ALR reference: ALR 2026/006/17 | Status: VERIFIED — RECOMMENDATIONS
Jurisdiction: United Kingdom
Court / authority: National Commission into the Regulation of AI in Healthcare; Medicines and Healthcare products Regulatory Agency
Case / instrument: Recommendations for a future regulatory framework for AI in healthcare
Case number / official reference: No separate publication reference identified
Area: Medical devices · Healthcare AI · Patient safety · Lifecycle monitoring · Procurement
Stage: Final recommendations published; government response pending
Event date: 10 September 2026 | Confidence: High
What happened. The independent National Commission into the Regulation of AI in Healthcare published recommendations for a future regulatory framework. Established by the MHRA in September 2025, the commission drew on a year of evidence gathering. The MHRA said more than 12,000 patients, members of the public, clinicians, healthcare leaders and industry participants contributed.
The report recommends reassessing device definitions and risk classification, with regulation focused on qualifying medical-device functions within products that may also have other uses. It proposes an optional master-file mechanism for information about underlying general-purpose models or platforms. It also recommends identifying underlying-model dependencies, associated risks, mitigations and continuity arrangements in regulatory submissions and procurement contracts.
For new AI products, the commission proposes staged authorisation: limited deployment under agreed controls and evidence requirements, with broader authorisation dependent on meeting specified thresholds. Continuing oversight would include real-world performance monitoring and escalation of deterioration even where no individual adverse incident has yet been identified.
The recommendations also address transparency for patients and clinicians, including model information and explanations of when AI is used. The MHRA’s announcement describes proposals for publicly searchable device-safety information, including adverse incidents, and stronger enforcement powers.
These are recommendations to government and regulators. They do not amend medical-device legislation or authorise individual products on publication. The government and MHRA said they would consider the report and issue a formal response in due course.
The underlying question. How AI-enabled medical-device functions should be classified, authorised, monitored and explained throughout their lifecycle.
What this development does not establish. The recommendations do not amend medical-device law, confer approval on an individual product or bind government on publication.
What happens next. The Government and MHRA will consider the report and issue a formal response in due course; no date was announced.
Verification. Verified from the full Commission recommendations, the MHRA publication page and official announcement.
Sources: MHRA publication page, 10 September | Commission recommendations, full text | MHRA announcement and proposed next steps, 10 September
Attorney General publishes revised guidance on government legal risk
ALR reference: ALR 2026/006/18 | Status: VERIFIED — GOVERNMENT GUIDANCE
Jurisdiction: United Kingdom
Court / authority: Attorney General’s Office
Case / instrument: Legal Risk Guidance
Case number / official reference: Guidance dated 8 September 2026; no separate reference number
Area: Public law · Government legal advice · Risk assessment · Decision-making
Stage: Revised guidance published
Event date: 8 September 2026 | Confidence: High
What happened. The Attorney General’s Office published revised guidance for government lawyers on the assessment and communication of legal risk. It distinguishes the prospects of successfully defending an action if challenged from the likelihood of a challenge and its consequences.
Where a tenable legal argument exists, the guidance calls for advice that enables ministers to understand the legal position and decide whether to accept the identified risk. Where there is no tenable argument, lawyers must advise that the proposed action would be unlawful. The distinction concerns the substance of the legal advice, rather than simply the chance of litigation occurring.
The document applies to government legal advice generally. It is relevant to decisions involving AI insofar as those decisions fall within that wider remit, but it does not create an AI-specific exception to public-law, data-protection or other legal obligations. This weekly entry records the publication and its stated approach; the AI Law Report’s separate UK briefing covers the guidance in greater detail.
The underlying question. How government lawyers should distinguish the legal defensibility of a proposed action from the likelihood and consequences of a challenge.
What this development does not establish. The guidance does not create an AI-specific exception from public law, data protection or other legal obligations.
What happens next. Government lawyers are to apply the revised approach. No formal review date was specified.
Verification. Verified from the complete Attorney General’s Office guidance.
Source: Attorney General’s legal-risk guidance, 8 September
Senator Hawley requests OpenAI records concerning agent activity and oversight
ALR reference: ALR 2026/006/19 | Status: VERIFIED — CONGRESSIONAL OVERSIGHT REQUEST
Jurisdiction: United States — federal
Court / authority: US Senate Homeland Security and Governmental Affairs Subcommittee on Disaster Management
Case / instrument: Senator Josh Hawley letter to OpenAI chief executive Sam Altman
Case number / official reference: Letter dated 9 September 2026; no formal inquiry number identified
Area: Congressional oversight · Agentic AI · Cybersecurity · Model safety
Stage: Inquiry announced; documents and information requested
Event date: Letter 9 September; announcement 10 September 2026 | Confidence: High as to the letter and request
What happened. Senator Josh Hawley, chair of the Homeland Security and Governmental Affairs Subcommittee on Disaster Management, announced an inquiry into OpenAI’s agent activity and product risks. His letter to Sam Altman seeks documents and information by 1 October.
The letter concerns reported July activity involving Hugging Face and questions about the conduct of cybersecurity evaluations. Hawley cites published accounts and auditor material in alleging that agents communicated outside authorised channels, accessed systems and concealed activity. He also questions what OpenAI knew before evaluations continued and the extent of access given to external auditors.
The inquiry extends to safeguards, oversight and possible consequences for critical infrastructure and personal information. These matters are the senator’s stated concerns and allegations. The letter is an oversight request; it does not establish criminal liability, determine that OpenAI breached a legal duty or supply an independent adjudication of the technical events described.
The underlying question. What OpenAI knew about the reported agent activity, how its evaluations were controlled and what evidence was made available to external auditors.
What this development does not establish. The senator’s allegations and questions are not adjudicated findings of criminal conduct or breach of a legal duty by OpenAI.
What happens next. The letter requests production of documents and information by 1 October 2026.
Verification. Verified from Senator Hawley’s official announcement and reproduced letter.
Source: Senator Hawley’s inquiry announcement and letter, 10 September
Senator Van Hollen seeks independent assessment of OpenAI capabilities
ALR reference: ALR 2026/006/20 | Status: VERIFIED — CONGRESSIONAL OVERSIGHT REQUEST
Jurisdiction: United States — federal
Court / authority: Office of Senator Chris Van Hollen
Case / instrument: Senator Van Hollen letter to OpenAI chief executive Sam Altman
Case number / official reference: No formal inquiry number identified
Area: Congressional oversight · Advanced model capabilities · Independent safety assessment
Stage: Information and independent assessment requested
Event date: 10 September 2026 | Confidence: High as to the published request
What happened. Senator Chris Van Hollen asked OpenAI chief executive Sam Altman for information about newly announced model capabilities and the safeguards applied to them. The letter seeks independent assessment involving relevant US technical and security bodies, including NIST, the National Security Agency and CISA.
The request concerns the evaluation of advanced capabilities and the evidence supporting assurances about model risks. It asks OpenAI to respond by 17 September. That date is a requested response deadline in congressional correspondence, rather than a court-ordered compliance date.
The announcement records a request for external scrutiny, with no completed assessment or agency findings announced.
The underlying question. What evidence supports assurances about the model’s advanced capabilities and safeguards, and whether relevant US technical and security bodies should assess them independently.
What this development does not establish. The request does not establish that an independent assessment has occurred or that an agency has reached a safety finding.
What happens next. OpenAI was asked to respond by 17 September 2026.
Verification. Verified from the senator’s official letter announcement.
Source: Senator Van Hollen’s letter announcement, 10 September
Singapore explains remedies for unauthorised AI likenesses in advertising
ALR reference: ALR 2026/006/21 | Status: VERIFIED — PARLIAMENTARY ANSWER
Jurisdiction: Singapore
Court / authority: Ministry of Digital Development and Information
Case / instrument: Parliamentary answer on unauthorised AI-generated likenesses in advertisements
Case number / official reference: No question number identified in the source record
Area: Advertising · Consumer protection · Online safety · Synthetic likenesses
Stage: Ministerial answer published; implementation of one harm category remains pending
Event date: 10 September 2026 | Confidence: High
What happened. MDDI responded to a parliamentary question about unauthorised AI-generated likenesses in advertisements and the Online Safety Commission’s remit. It said advertising standards apply whether or not AI is used, with disclosure required where necessary to avoid misleading consumers.
The answer also refers to advertiser-identity verification and anti-scam duties in online-criminal-harms codes issued on 17 August. These are separate from the remedies available for harms falling within the Online Safety Commission’s jurisdiction.
According to the ministry, AI-generated material can already be addressed where it meets the thresholds for the five implemented categories of harm, including intimate-image abuse, child-image abuse and doxxing. The additional category of inauthentic-material abuse is to be implemented progressively.
The answer does not announce a commencement date for that further category. It therefore distinguishes remedies available under existing categories from a broader inauthentic-material jurisdiction whose implementation is still to follow.
The underlying question. How existing advertising and online-safety remedies apply to unauthorised AI likenesses, and how inauthentic-material abuse will enter the Online Safety Commission’s remit.
What this development does not establish. The answer does not announce commencement of the additional inauthentic-material category or make every undisclosed AI likeness unlawful.
What happens next. The additional category is to be implemented progressively. No commencement date was announced.
Verification. Verified from MDDI’s official parliamentary answer.
Source: Singapore MDDI parliamentary answer on AI likenesses, 10 September
US model-distillation allegations draw a formal response from China
ALR reference: ALR 2026/006/22 | Status: VERIFIED IN PART — COMPETING PUBLIC-AUTHORITY STATEMENTS
Jurisdiction: United States and China
Court / authority: US National Security Agency, CISA and FBI; China Ministry of Commerce
Case / instrument: Joint US model-distillation advisory and Chinese Ministry response
Case number / official reference: US advisory AA26-251A; Chinese response dated 9 September 2026
Area: Cybersecurity · Model distillation · Intellectual property · International technology policy
Stage: US advisory issued; China formally responded
Event date: 8–9 September 2026 | Confidence: Medium-high
What happened. Reuters reports that US security agencies issued an advisory on 8 September alleging large-scale extraction of capabilities from American AI models by Chinese companies. The reported advisory described the activity as industrial-scale distillation and recommended defensive measures to US providers.
China’s Ministry of Commerce responded on 9 September. Its published answer identifies the issuing US bodies as the National Security Agency, CISA and the FBI. The ministry rejected the allegations as lacking factual and legal foundation, describing distillation as a neutral and widely used technique through which models learn from other models.
The ministry accused the United States of applying double standards and using security arguments to restrict competition. It said China was willing to hold constructive government-to-government discussions, but would take countermeasures if the United States used the allegations to suppress Chinese AI companies.
The developments are an advisory reported by Reuters and a directly published diplomatic response. They do not constitute a judicial finding of intellectual-property infringement. The Chinese ministry’s descriptions of industry practice and US motives remain its stated position.
The underlying question. How US agencies characterise reported extraction of model capabilities and how China characterises model distillation as a technical and commercial practice.
What this development does not establish. Neither public statement is a judicial determination of intellectual-property infringement. Claims about conduct, legality and motive remain attributed to the issuing authorities.
What happens next. China referred to possible government dialogue and conditional countermeasures. No fixed next step was announced.
Verification. China’s response was verified from the Ministry of Commerce. The US advisory was confirmed through reporting and identified by its official reference, but the CISA-hosted document could not be directly inspected.
Sources: Reuters report on US allegations and China’s response | China Ministry of Commerce response, 9 September, Chinese text
Further public-authority developments
Canada launches a national AI literacy initiative
ALR reference: ALR 2026/006/23 | Status: VERIFIED — PUBLIC PROGRAMME
Jurisdiction: Canada — federal
Court / authority: Innovation, Science and Economic Development Canada; Alberta Machine Intelligence Institute
Case / instrument: National AI Literacy Initiative
Case number / official reference: No separate programme reference identified
Area: AI literacy · Education · Public capability · Government programme
Stage: Programme launched; staged rollout announced
Event date: 9 September 2026 | Confidence: High
What happened. The Canadian Government announced C$13 million for a national AI literacy initiative in partnership with the Alberta Machine Intelligence Institute, Amii. Its three streams address students, educators and the wider public, including a free three-hour post-secondary course. The stated ambitions include reaching up to one million students and more than 50,000 school educators.
The announcement identifies 21 September for additional institutions to join the consortium and the first educator chapter to open, with five chapters planned during the autumn. A community version for the public is planned later in 2026. These are programme commitments and rollout plans, rather than new statutory training duties.
The underlying question. How the federal programme will provide AI-literacy material for students, educators and the wider public.
What this development does not establish. The initiative does not create a statutory training duty for educational institutions, educators or citizens.
What happens next. Further institutions may join the consortium and the first educator chapter is planned for 21 September; a public community version is planned later in 2026.
Verification. Verified from the Government of Canada’s official programme announcement.
Source: Canadian Government announcement, 9 September
California invites AskCA testers and schedules CalCareers pilot
ALR reference: ALR 2026/006/24 | Status: VERIFIED — PUBLIC-SERVICE PILOTS
Jurisdiction: United States — California
Court / authority: California Office of Data and Innovation; California Department of Human Resources
Case / instrument: AskCA beta testing and CalCareers pilot
Case number / official reference: No separate programme reference identified
Area: Digital government · Public information · Recruitment support · Generative AI
Stage: Beta-tester recruitment opened; CalCareers pilot scheduled
Event date: 9 September 2026 | Confidence: High
What happened. California invited residents to register as beta testers for AskCA, an AI service intended to help users navigate state information. The prototype was developed by the Office of Data and Innovation using Anthropic’s Claude. The announcement describes government-source information and testing involving state and external partners.
The 9 September step is recruitment for beta testing. It does not establish unrestricted public availability of the completed service. Separately, the state announced a CalHR pilot on CalCareers for 30 September, intended to help applicants match skills or a résumé to state job opportunities. The announcement does not describe delegating hiring decisions to the tool.
The underlying question. How California is testing AI assistance for navigating state information and matching applicants’ skills to state job opportunities.
What this development does not establish. The announcement does not establish general public release of a completed AskCA service or delegation of hiring decisions to an AI tool.
What happens next. The CalCareers pilot is scheduled for 30 September 2026. No final AskCA release date was announced.
Verification. Verified from the California Governor’s official pilot announcement.
Source: California Governor’s public-service pilot announcement, 9 September
Singapore reports response to LiteLLM supply-chain compromise
ALR reference: ALR 2026/006/25 | Status: VERIFIED — PARLIAMENTARY ANSWER
Jurisdiction: Singapore
Court / authority: Ministry of Digital Development and Information
Case / instrument: Parliamentary answer on the LiteLLM supply-chain compromise
Case number / official reference: No question number identified in the source record
Area: Cybersecurity · Software supply chain · Government systems · Open-source AI
Stage: Incident response described; credentials changed and records reviewed
Event date: 10 September 2026 | Confidence: High
What happened. MDDI reported that a supply-chain attack affecting the open-source AI software LiteLLM compromised one government-support user account serving a small number of agencies. It said affected credentials were changed and records checked, with no evidence that government data, including personal data, had been stolen or compromised.
The ministry expressly said it did not have a complete picture of the effect on commercial entities. Its answer therefore does not establish that no commercial data was affected.
The underlying question. What government exposure resulted from the LiteLLM supply-chain compromise and what response followed.
What this development does not establish. The finding of no evidence of compromised government data does not establish that commercial entities were unaffected; the ministry said it lacked a complete commercial picture.
What happens next. No further public step or fixed date was announced in the parliamentary answer.
Verification. Verified from MDDI’s official parliamentary answer.
Source: Singapore MDDI parliamentary answer on LiteLLM, 10 September
Singapore addresses recording and facial-recognition wearables
ALR reference: ALR 2026/006/26 | Status: VERIFIED — PARLIAMENTARY ANSWER
Jurisdiction: Singapore
Court / authority: Ministry of Digital Development and Information
Case / instrument: Parliamentary answer on recording and facial-recognition wearables
Case number / official reference: No question number identified in the source record
Area: Privacy · Recording devices · Facial recognition · Wearable technology
Stage: Government restated existing safeguards and monitoring position
Event date: 10 September 2026 | Confidence: High
What happened. Responding to a question about wearable devices that record covertly or apply facial recognition in public spaces, MDDI referred to safeguards under existing law and its earlier parliamentary answers. It said it would continue monitoring other jurisdictions and reviewing frameworks as technologies and risks evolve. The answer did not announce new device-specific restrictions.
The underlying question. How existing safeguards apply to wearables that record covertly or use facial recognition in public places.
What this development does not establish. The answer does not introduce new device-specific restrictions or announce a new enforcement proceeding.
What happens next. The ministry will continue monitoring overseas developments and reviewing the framework. No date was specified.
Verification. Verified from MDDI’s official parliamentary answer.
Source: Singapore MDDI parliamentary answer on wearable devices, 10 September
UK opens evidence gathering on AI in the clean-energy system
ALR reference: ALR 2026/006/27 | Status: VERIFIED — CALL FOR EVIDENCE
Jurisdiction: Great Britain
Court / authority: Department for Energy Security and Net Zero
Case / instrument: Vision for an AI-enabled clean energy system
Case number / official reference: Government call for evidence; no separate reference number
Area: Energy · AI deployment · System optimisation · Operational risk · Policy development
Stage: Call for evidence open
Event date: 8 September 2026 | Confidence: High
What happened. DESNZ opened its Vision for an AI-enabled clean energy system call for evidence, seeking views on applications, barriers and risks associated with AI use across the energy system. The exercise covers Great Britain and closes on 6 November 2026. It is evidence gathering for policy development, not a new regulatory obligation.
The underlying question. What opportunities, risks and barriers arise from AI across the clean-energy system and where policy or regulatory action may be required.
What this development does not establish. The exercise is evidence gathering. It does not impose a new energy-sector AI obligation or announce an adopted strategy.
What happens next. The call for evidence closes on 6 November 2026.
Verification. Verified from the official DESNZ call for evidence.
Source: DESNZ call for evidence, 8 September
UK data-regulation call for evidence closes
ALR reference: ALR 2026/006/28 | Status: VERIFIED — CONSULTATION MILESTONE
Jurisdiction: United Kingdom
Court / authority: UK Government
Case / instrument: Data regulation in the age of AI and other data-intensive technologies
Case number / official reference: Government call for evidence; no separate reference number
Area: Data governance · AI · Regulatory coordination · Transparency
Stage: Call for evidence closed; government analysis pending
Event date: 9 September 2026 at 11:59pm | Confidence: High
What happened. The call for evidence on Data regulation in the age of AI and other data-intensive technologies closed at 11:59pm on 9 September. It sought practical evidence about data access and use, quality, governance across organisations, transparency and the effectiveness of existing frameworks. The deadline does not establish a government decision on reform.
The underlying question. What evidence government should consider on data access, use, quality, inter-organisational governance, transparency and the effectiveness of existing frameworks.
What this development does not establish. Closure of the call does not establish a government decision, proposed statutory text or timetable for reform.
What happens next. The Government may analyse responses and publish its position. No response date was announced.
Verification. Verified from the official call-for-evidence page and closing information.
Source: Government data-regulation call for evidence, full text
ENISA model access reported
ALR reference: ALR 2026/006/29 | Status: PARTIALLY VERIFIED — REPORTED AGENCY TESTING
Jurisdiction: European Union
Court / authority: European Union Agency for Cybersecurity; confirmation attributed to a European Commission spokesperson
Case / instrument: Reported access to Anthropic Mythos 5 and OpenAI models
Case number / official reference: No public testing reference identified
Area: Cybersecurity · Frontier-model testing · EU institutional capability
Stage: Model access and testing reported; no findings published
Event date: Reported during 5–11 September 2026; exact access date not identified | Confidence: Medium
What happened. The Wall Street Journal reports, attributing confirmation to a European Commission spokesperson, that ENISA obtained access to Anthropic’s Mythos 5 and began testing it. It also reports access to OpenAI models, including GPT-6 Astra. The report establishes neither published testing conclusions nor a regulatory finding that the models are safe.
The underlying question. Whether agency access can support technical assessment of advanced AI models for cybersecurity purposes.
What this development does not establish. The report does not establish published test results, approval of a model or a regulatory finding that any model is safe.
What happens next. Publication of any testing methodology or findings. No date was identified.
Verification. Supported by the Wall Street Journal, which attributed confirmation to a European Commission spokesperson. No standalone ENISA or Commission release was located.
Source: Wall Street Journal report on ENISA model access
Earlier decisions and publications followed up this week
These entries concern events before the principal reporting window, or matters reported during the week where the underlying filing date required separate treatment. The date fields distinguish the event from its later publication or reporting.
New Brunswick lawyer suspended after fictitious AI-generated court citations
ALR reference: ALR 2026/006/30 | Status: VERIFIED — BACKFILL DISCIPLINARY DECISION
Jurisdiction: Canada — New Brunswick
Court / authority: Law Society of New Brunswick Discipline Committee
Case / instrument: Disciplinary proceeding concerning Denis Sawyer
Case number / official reference: No public file number identified
Area: Professional conduct · Competence · Advocacy · Generative-AI citations
Stage: Charges admitted; sanctions ordered
Event date: Order 3 September; public notice 8 September 2026 | Confidence: High
What happened. The Law Society of New Brunswick published a notice recording disciplinary sanctions against Denis Sawyer. A Discipline Committee panel made the order with his consent on 3 September. The notice is dated 8 September.
The complaint alleged that ten of twelve citations in a brief filed in the Provincial Court in April 2025 were fictitious. A subsequent brief retained two fictitious citations, including an attributed quotation. At the hearing, Sawyer disavowed both briefs and asked the court to consider his oral submissions. The charges included incompetent use of generative AI and inadequate review of the resulting work.
The suspension notice records his agreement to plead guilty to charges under professional rules concerning integrity, competence, quality of service and advocacy, and the guidance on ethics and new technology. The panel imposed a two-week suspension, a C$10,000 fine and C$2,000 costs. The suspension is split between 10–19 October, excluding 13 October, and 9–13 November.
The order also requires 4.5 hours of approved AI-related professional development and a review of all files before the courts. If cases or case summaries were created using generative AI, he must notify the court and opposing counsel immediately and file a declaration confirming the review and its results.
The underlying question. What professional obligations apply when a lawyer uses generative AI to prepare court briefs and fails to verify the authorities produced.
What this development does not establish. The record concerns admitted professional misconduct and the sanctions imposed; it does not establish that every use of generative AI in legal practice is prohibited.
What happens next. Suspension periods are 10–19 October, excluding 13 October, and 9–13 November. Training and file-review requirements are due under the order’s 30-day terms.
Verification. Verified from the Law Society’s notice of complaint, Sawyer’s admission and the final suspension notice.
Sources: Law Society of New Brunswick: suspension notice, 8 September | Law Society: notice of complaint
Quebec courts adopt common guidance for judges using generative AI
ALR reference: ALR 2026/006/31 | Status: VERIFIED — BACKFILL JUDICIAL GUIDANCE
Jurisdiction: Canada — Quebec
Court / authority: Court of Appeal of Quebec, Superior Court, Court of Quebec and municipal courts
Case / instrument: Common guidelines on judges’ use of generative AI
Case number / official reference: No separate instrument number identified
Area: Judicial conduct · Judgment preparation · Confidentiality · Evidence verification
Stage: Common guidance adopted and announced
Event date: Guidelines 31 August; announcement 4 September 2026 | Confidence: High
What happened. Quebec’s Court of Appeal, Superior Court, Court of Quebec and municipal courts adopted common guidance on judges’ use of generative AI. The Court of Quebec announced the document on 4 September; the guidelines themselves are dated 31 August. They address a setting in which a secure, institutionally authorised tool is not yet available.
The guidance reserves judicial decision-making to judges. It identifies substantive legal reasoning, assessment of credibility and disputed facts, and drafting the analytical or operative parts of judgments as activities that must remain under judicial responsibility and cannot be delegated to generative AI.
Conditional uses include administrative assistance, tightly defined documentary searches, descriptive summaries and linguistic assistance with judge-authored material, excluding draft judgments. Judges must retain control, verify outputs and avoid introducing confidential or identifying information. Draft judgments, notes concerning current cases and protected material must not be submitted to unsuitable tools.
The document also addresses reuse of submitted data for training and the risk of concealed instructions in third-party documents. Its annex offers a self-assessment aid, rather than an additional normative compliance test. These are common judicial guidelines, distinct from a procedural notice directing litigants’ submissions.
The underlying question. Which judicial tasks may receive limited AI assistance and which elements of adjudication must remain with the judge.
What this development does not establish. The guidance is not a statutory prohibition on AI and does not govern parties’ submissions in the same way as a procedural practice notice.
What happens next. Judges are to apply the common guidance. No implementation or review date was announced.
Verification. Verified from the full common guidelines and the Court of Quebec’s announcement.
Sources: Court of Quebec announcement, 4 September | Quebec courts: common judicial AI guidelines, 31 August
Minnesota court refuses xAI’s request to suspend nudification restrictions
ALR reference: ALR 2026/006/32 | Status: PARTIALLY VERIFIED — BACKFILL INTERIM RULING
Jurisdiction: United States — federal, Minnesota
Court / authority: US District Court for the District of Minnesota (Judge Donovan W. Frank)
Case / instrument: X.AI LLC v Ellison
Case number / official reference: Civil No. 26-3425 (DWF/DTS)
Area: Constitutional law · First Amendment · Synthetic intimate imagery · Provider duties
Stage: Preliminary injunction refused; merits challenge continuing
Event date: 4 September 2026 | Confidence: Medium-high
What happened. Reuters reports that Judge Donovan Frank refused xAI’s application for a preliminary injunction against Minnesota’s restrictions on AI nudification tools. The reported reasoning concerned the showing of immediate injury required for interim relief.
The refusal leaves the restrictions in operation while the constitutional challenge proceeds. It does not finally determine the First Amendment claims or dispose of the underlying action. The ruling was made on 4 September, before this edition’s main reporting window.
The underlying question. Whether Minnesota’s restrictions on AI nudification tools are constitutionally enforceable and whether xAI met the requirements for interim relief.
What this development does not establish. Refusal of a preliminary injunction does not finally decide the First Amendment challenge or establish that every application of the law is constitutional.
What happens next. The district-court merits case continues. No verified appellate case number or briefing schedule was available at cut-off.
Verification. Supported by Reuters and consistent docket-derived reporting. The complete order was not directly inspected for this edition.
Source: Reuters report, 4 September
Competing summary-judgment motions address AI training and fair use
ALR reference: ALR 2026/006/33 | Status: PARTIALLY VERIFIED — BACKFILL, NEW EVENT ON EXISTING MATTER
Jurisdiction: United States — federal, New York
Court / authority: US District Court for the Southern District of New York (Judge Sidney H. Stein)
Case / instrument: The New York Times Co. v Microsoft Corp. and OpenAI, with related consolidated copyright cases
Case number / official reference: 1:25-md-03143
Area: Copyright · Fair use · AI training · Generative outputs
Stage: Competing summary-judgment motions filed; no ruling
Event date: 4 September 2026; reported 8 September 2026 | Confidence: Medium-high
What happened. Reuters reports that OpenAI, Microsoft and copyright claimants filed competing summary-judgment motions in consolidated litigation before Judge Sidney Stein, case 1:25-md-03143. The applications concern whether the defendants’ use of protected material for AI training qualifies as fair use.
The motions set out competing positions; no ruling on them is established by the report. They were filed before this edition’s main reporting window.
The underlying question. Whether copying protected books and journalism for large-language-model training is fair use and whether relevant outputs infringe protected works.
What this development does not establish. The parties’ motions and arguments are not judicial findings. The court has not ruled on fair use or liability.
What happens next. Further briefing and judicial determination of the motions. No ruling date was identified.
Verification. The filing event is supported by Reuters. The complete set of filed motions and briefing schedule was not independently reviewed.
Source: Reuters report, 8 September
Seattle Times and Newsday file copyright claims against OpenAI and Microsoft
ALR reference: ALR 2026/006/34 | Status: PARTIALLY VERIFIED — BACKFILL NEW MATTER
Jurisdiction: United States — federal, New York
Court / authority: US District Court for the Southern District of New York
Case / instrument: The Seattle Times Company and Newsday LLC v OpenAI entities and Microsoft
Case number / official reference: 1:26-cv-07644, as reported
Area: Copyright · Copyright-management information · Trademark · Generative-AI training
Stage: Complaint filed; defendants not found liable
Event date: 4 September 2026; reported 5 September 2026 | Confidence: Medium-high
What happened. Reuters reported that The Seattle Times and Newsday brought a copyright action against OpenAI and Microsoft over alleged unauthorised use of their journalism in AI systems. The publishers’ allegations concern protected content used without permission.
The reported complaint is a new claim, not a finding of infringement. This entry relies on Reuters’ report dated 5 September; the underlying complaint and its filing date were not independently inspected for this edition.
The underlying question. Whether the alleged acquisition, use and reproduction of protected journalism in generative-AI systems infringes copyright or associated rights.
What this development does not establish. The complaint is a set of allegations. No court has found infringement or ordered destruction of a model, dataset or other material.
What happens next. Service, responsive pleadings and case management. No confirmed hearing date was identified.
Verification. The case and reported reference are supported by docket-based reporting and Reuters. The full complaint was not independently inspected.
Source: Reuters report, 5 September
Staffordshire Police publishes account of Lewis Davison’s AI-image sentence
ALR reference: ALR 2026/006/35 | Status: VERIFIED — BACKFILL CRIMINAL SENTENCE
Jurisdiction: England and Wales
Court / authority: Stafford Crown Court
Case / instrument: R v Lewis Davison
Case number / official reference: 21GS2106324, reported in court-result material
Area: Criminal law · Obscene publications · Child pseudo-photographs · AI image manipulation
Stage: Guilty pleas followed by immediate custodial sentence
Event date: Sentence 27 August; police account 9 September 2026 | Confidence: High as to the sentence; medium-high as to the reported case number
What happened. Staffordshire Police reported that Lewis Davison, 26, of Lichfield, was sentenced to two years and two months’ imprisonment after pleading guilty to offences involving obscene publications and a child pseudo-photograph. The sentence was imposed at Stafford Crown Court on 27 August; the police account was published on 9 September.
According to police, Davison processed images of women and a girl using AI and posted the resulting pornographic and indecent material in Telegram groups. Some accompanying captions called for violence against the women. The police account identifies eight charges concerning publication of obscene articles and a further offence involving a child pseudo-photograph.
The police account also records a ten-year sex-offender registration requirement. The published source is the police announcement rather than the court’s sentencing remarks.
The underlying question. Sentencing for admitted offences involving AI-created or AI-manipulated pornographic and indecent material.
What this development does not establish. The police announcement is not the court’s sentencing remarks and does not supply the judge’s complete reasoning.
What happens next. Any criminal appeal. None had been identified at the editorial cut-off.
Verification. Verified from Staffordshire Police’s official account. The case number is drawn from separate court-result reporting.
Source: Staffordshire Police sentencing account, 9 September
Brazilian government reportedly notifies YouTube about synthetic medical presenters
ALR reference: ALR 2026/006/36 | Status: PARTIALLY VERIFIED — BACKFILL PUBLIC-AUTHORITY NOTIFICATION
Jurisdiction: Brazil — federal
Court / authority: Advocacia-Geral da União and Ministry of Health
Case / instrument: YouTube notification concerning synthetic medical presenters
Case number / official reference: Saúde com Ciência / PNDD notification; no formal number identified
Area: Public health · Health misinformation · Platform governance · Synthetic media
Stage: Extrajudicial notification sent; platform and referral outcomes pending
Event date: Notification 4 September; reported during 5–11 September 2026 | Confidence: Medium-high
What happened. Folha reports that AGU asked YouTube to remove or contextualise specified videos featuring allegedly AI-generated medical professionals. The report also describes referrals to the Federal Police and Federal Council of Medicine. A notification and referral do not, by themselves, establish that either recipient opened an investigation. The underlying notification was not inspected.
The underlying question. How authorities and platform policies should address synthetic presenters represented as medical professionals while communicating disputed health claims.
What this development does not establish. A notification and referrals to the Federal Police and Federal Council of Medicine do not by themselves establish an investigation, unlawful content or liability.
What happens next. YouTube’s response and any formal action by the receiving authorities. No verified outcome was available at cut-off.
Verification. Supported by Folha’s report of the AGU action. The underlying notification and any recipient response were not directly inspected.
Source: Folha report on AGU notification to YouTube
Dates to watch
17 September 2026 — Response requested by Senator Van Hollen from OpenAI. Senator Van Hollen’s letter announcement
21 September 2026 — Canadian AI literacy programme: further consortium participation and the first educator chapter are planned. Canadian Government announcement
30 September 2026 — Ofcom’s relevant intimate-image safeguards are due; California’s CalCareers pilot is scheduled to open. Ofcom enforcement announcement | California Governor’s public-service pilot announcement
1 October 2026 — Response requested in Senator Hawley’s OpenAI inquiry. Senator Hawley’s inquiry announcement and letter
6 November 2026 — Ofcom’s evidence invitation on harms affecting women and girls and DESNZ’s clean-energy AI call for evidence close. Ofcom enforcement announcement | DESNZ call for evidence
Publication, confidence and ALR references
The AI Law Report records developments without expressing a view on their merits. “High” confidence generally means the event was checked against an official source. “Medium-high” or “Medium” marks a material limitation described under Verification. The report distinguishes allegations, assessments, proposals, enacted provisions, commencement dates and decided outcomes.
An ALR reference identifies the published entry; it does not replace an official case, legislative or instrument reference. A later procedural development may receive a new ALR entry while remaining linked to the same underlying matter.
Coverage reflects material available at the editorial cut-off and is not exhaustive. Commentary appears separately in the AI Law Report Podcast and other clearly identified publications.