AI Compliance Map — Aug 21–28, 2026: New Zealand proposes platform controls as California advances workplace and health AI bills
Six verified AI regulatory actions across California, New Zealand, China, and Australia, with status, scope, deadlines, and compliance impact for multinational teams.
The exact coverage window is 21 August 2026 at 17:00 through 28 August 2026 at 17:00, UTC-05:00. Six AI-related regulatory actions met the date and evidence test for this edition: three California bills, one New Zealand bill, one Chinese ethics guidance notice, and one Australian tribunal guidance release.
The immediate work is mostly preparation and monitoring. The Fair Work Commission's GenAI case guidance will apply on 20 October. California's measures remain at different legislative stages, and New Zealand's platform bill remains a proposal. China's medical-imaging guidance is expressly offered for reference. The legal status beside each item matters as much as the subject: binding or scheduled guidance calls for implementation; a bill calls for impact assessment and legislative monitoring; reference guidance calls for a voluntary control review.
Action list
- Prepare Fair Work Commission case controls before 20 October. Any party using generative AI to prepare Commission documents will need a process to disclose when and how it used GenAI, verify accuracy and relevance, and protect the first-person basis of witness statements and declarations. 1
- Put California public-sector AI procurement on a pre-deployment watchlist. Enrolled AB 2656 would create a 45-day notice step for covered public employers before they develop, purchase, or require GenAI for work within a represented job classification. The bill is enrolled and has passed both chambers, while the record reviewed here shows no enactment date. 2
- Ask California higher-education vendors for procurement evidence early. AB 2392 is an amended proposal that would require the California Community Colleges and California State University to develop responsible GenAI procurement standards and training, with reporting within 60 days after a systemwide contract until standards are adopted. Vendors serving those segments should be ready to answer questions about risk assessments, data practices, incident reporting, training data, and user training if the proposal advances. 3
- Monitor New Zealand's age-assurance and child-safety design requirements as a legislative proposal. Government Bill 339 would cover certain New Zealand-accessible platforms, including services that use AI primarily to simulate a social or emotional connection. The bill would require reasonable steps to prevent under-16 accounts and written child-safety risk assessments; its proposed requirements would begin six months after Royal assent. 4
- Keep clinical decision-support bias controls on California's legislative radar. SB 503 remains in progress after Assembly amendments were concurred in and the bill was ordered to engrossing and enrolling on 25 August. The proposal would require developers and deployers of clinical decision-support systems to identify, document, monitor, and mitigate known or reasonably foreseeable biased impacts. 5
Scan table
| Date | Jurisdiction and actor | Action and status | Affected scope | Compliance impact |
|---|---|---|---|---|
| 21 Aug | California Legislature | AB 2392 amended in the Senate; proposed | California Community Colleges, California State University, and requested participation by the University of California | Higher-education GenAI procurement, training, annual review, four-year training records, and interim contract reporting would be required if enacted. 3 |
| 24 Aug | Fair Work Commission, Australia | GenAI case guidance released; scheduled to apply 20 Oct | Parties using GenAI to prepare documents in Commission cases | Add disclosure, accuracy, relevance, and witness-authenticity checks to case-preparation workflows before the application date. 1 |
| 24 Aug | New Zealand Government | Online Safety Bill 339 introduced; proposed | Operators of certain age-restricted platforms accessible in New Zealand, including specified social AI companion services | Prepare for proposed under-16 account controls and written child-safety risk assessments; commencement depends on Royal assent and would follow six months later. 4 |
| 25 Aug | California Legislature | AB 2656 passed the Assembly; later enrolled 27 Aug; pending enactment | Certain state and local public employers and recognized employee organizations | A covered employer would give at least 45 days' written notice before specified GenAI activity affecting a represented job classification. 2 |
| 25 Aug | California Legislature | SB 503 Assembly amendments concurred in; in progress | Developers and deployers of clinical decision-support systems used in health programs or activities | The proposal would add intended-use and risk statements, developer documentation, monitoring, and reasonable, proportionate bias mitigation. 5 |
| 27 Aug | China's Ministry of Science and Technology | Medical-imaging AI ethics guidance announced; voluntary/reference guidance | Relevant research institutions and researchers conducting AI medical-imaging research | Use the guidance as a reference point for internal research-ethics review; the notice describes it as material for reference rather than a binding rule. 6 |
United States: California moves through several legislative stages
California supplied all three verified US entries this week. They reach different parts of the AI supply chain, so a single "California AI bill" response would miss the operational distinction between public employment, higher education, and clinical decision support.
AB 2656: public-employer notice before represented work changes
The California Legislature's official bill record shows AB 2656 enrolled on 27 August after Senate passage on 24 August and Assembly passage on 25 August. The bill remains enrolled rather than enacted in the record reviewed for this issue. 2
The proposal would require certain state and local public employers to give a recognized employee organization at least 45 days' written notice before developing, purchasing, or requiring the use of GenAI to perform a service within the scope of work of the represented job classification. 2
Compliance impact: public-sector employers and vendors should identify deployments that could perform represented work and map the notice trigger into procurement and change-management calendars. The 45-day step is a proposed pre-action control until the bill becomes law.
AB 2392: procurement standards and training for public higher education
The official record shows AB 2392 amended in the Senate on 21 August. The measure is a proposal concerning GenAI procurement standards and training for California's public postsecondary education segments. 3
The California Community Colleges and California State University would jointly convene an intersegmental working group. The University of California's participation is requested. The proposed working group would develop procurement standards covering documented risk assessment, harmful-output safeguards, safety monitoring, incident reporting, data use and deletion policies, data minimization, training-data disclosure, and certain vendor practices. 3
The proposal would also require training for applicable students, faculty, and staff. The training would be reviewed at least once per academic year, and records would be retained for at least four years after completion. Recommendations would be presented on or before 1 January 2028. Until procurement standards are adopted, a systemwide GenAI contract would trigger a written report within 60 days for the California Community Colleges and California State University. 3
Compliance impact: vendors serving these institutions should maintain a package that can answer the proposed procurement questions: risk assessment, privacy and retention, safety monitoring, incident response, training-data provenance, and user training. Those controls are preparation for a pending bill, rather than current statewide obligations under this entry.
SB 503: clinical decision-support bias controls remain pending
California Digital Democracy lists SB 503's latest version date as 27 August and its status as "In Progress." The record says Assembly amendments were concurred in on 25 August and the bill was ordered to engrossing and enrolling. This is a secondary legislative record, so the status should be treated as a pending proposal rather than enacted law. 5
SB 503 concerns clinical decision-support systems: AI systems that produce a prediction, classification, recommendation, evaluation, or analysis that aids diagnosis or treatment decisions. The proposal would require developers to identify known or reasonably foreseeable bias risks, publish intended-use and risk statements, and provide documentation. Developers would make reasonable efforts to mitigate those risks, while deployers would regularly monitor systems and take reasonable and proportionate mitigation steps. 5
Compliance impact: clinical AI teams should keep model cards, intended-use boundaries, bias testing, monitoring results, and mitigation records ready for review. SB 503 supplies a legislative signal for those records; this issue assigns it no effective date because the bill remains pending.
Europe and the UK: no qualifying in-window entry verified
No qualifying AI regulatory action from the European Union or the United Kingdom was verified for the stated window. Older AI Act deadlines, general guidance, and pages without a governing action date are outside this issue's admission test. The absence of an entry here is a coverage result for this edition, not a finding that no EU or UK activity occurred.
China: medical-imaging ethics guidance for reference
China's Ministry of Science and Technology announced the Guidelines for Ethics in Artificial Intelligence Medical Imaging Research on 27 August at 17:00. The notice says the National Committee of Experts on Science and Technology Ethics' medical ethics subcommittee prepared the guidance to support responsible innovation and healthy development in AI medical-imaging research. 6
The notice says relevant research institutions and researchers may use the guidance for reference. The announcement therefore establishes voluntary/reference status in this digest; it supplies no basis here for treating the document as a binding national rule. 6
Compliance impact: medical-imaging research organizations can compare their ethics-review, documentation, and governance practices with the guidance. The notice's public summary does not provide enough detail to attribute specific control requirements to the document, so this issue records the release and its status without adding unverified provisions.
Asia-Pacific: New Zealand's platform bill and Australia's tribunal guidance
New Zealand Bill 339: age-restricted platforms and social AI companions
New Zealand's official legislation record identifies Government Bill 339-1 as introduced on 24 August 2026. The bill would apply to an operator of an age-restricted platform when the platform enables access by people in New Zealand, regardless of whether the operator is inside or outside New Zealand. 4
The proposed definition includes certain services that exchange or display digital content and use AI solely or primarily to simulate a social, emotional, or other personal connection with a user. The bill would require reasonable steps to prevent a New Zealand person under 16 from having an account. It would also require a written child-safety risk assessment covering risks from content, platform design and operation, and advertising. 4
For an operator already covered when the proposed law commenced, the initial risk assessment would be due within six months after commencement. Ordinary assessments would be due within 12 months after a platform becomes age-restricted, with annual reviews and earlier reviews possible on regulatory direction. The platform duties would commence six months after Royal assent if the bill is enacted. 4
The proposed enforcement toolkit includes regulator warnings, enforceable undertakings, court applications for service-restriction and access-restriction orders, pecuniary penalties, and injunctions. Tier 1 penalties would reach NZD 500,000 for an individual or, for another operator, the greater of NZD 40 million or 10% of relevant global turnover for each financial year in which the act occurred. 4
Compliance impact: operators of social AI companion services and other potentially captured platforms should map account-age controls, child-safety risk assessment ownership, change-trigger reviews, and data handling for age checks. Those are readiness steps for a bill, not current New Zealand duties.
Australia: Fair Work Commission case guidance
The Fair Work Commission published its GenAI case guidance on 24 August and said the guidance note will apply from 20 October 2026. The guidance addresses parties using GenAI to prepare documents in Commission cases. 1
The guidance sets three requirements: parties must tell the Commission when and how they used GenAI; check that the document is correct and relevant to the case; and, for a witness statement or declaration, check that the document rests on the person's own knowledge, uses the person's own words, and is true to the best of that person's knowledge. The Commission also said its application and response forms will be updated by 20 October. 1
Compliance impact: legal and employee-relations teams should add GenAI disclosure and human-verification steps to filing workflows, preserve the source of witness evidence, and update templates before 20 October. The control applies to Commission-case use described by the guidance, rather than to every workplace use of GenAI.
Near-term deadline register
- 20 October 2026 — Fair Work Commission guidance applies. Case teams should have disclosure, accuracy, relevance, and witness-authenticity checks in place before this date. 1
- Before any covered California public-employer action — AB 2656's proposed 45-day notice. The notice would precede specified GenAI development, purchase, or required use affecting represented work if the enrolled bill becomes law. 2
- 1 January 2028 — AB 2392's proposed recommendations date. The date belongs to the proposal's working-group recommendations and becomes an operative planning deadline only if the bill is enacted in a form that retains it. 3
- Six months after Royal assent — New Zealand Bill 339's proposed commencement. The bill sets the interval; the actual date is unavailable while the measure remains a proposal. 4
- Six months after commencement — initial New Zealand risk assessment for existing operators, if enacted. Later reviews would recur annually, subject to the bill's proposed earlier-review power. 4
Coverage note
This issue contains six verified in-window actions. The entries cover California, New Zealand, China, and Australia. Research for this window verified no qualifying US federal, EU, UK, court-ruling, or regulatory-enforcement item for inclusion. The six entries do not establish that no other action occurred worldwide; they show the actions that met the available source, scope, and timestamp checks for this edition.
The practical split is clear from the statuses: one scheduled procedural requirement in Australia, three California proposals at amended, in-progress, or enrolled stages, one New Zealand proposal with potentially broad platform scope, and one Chinese guidance notice offered for reference. Policy teams can therefore separate work due before a stated date from controls that belong on legislative watchlists.
References
- 1
- 2AB 2656 bill text and status — California Legislature
leginfo.legislature.ca.gov
- 3AB 2392 bill text and status — California Legislature
leginfo.legislature.ca.gov
- 4
- 5SB 503: Health care services: artificial intelligence — California Digital Democracy
calmatters.digitaldemocracy.org
- 6
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.
Related content
More from this channel
- AI Compliance Map — Sep 18–25, 2026: Australia investigates an AI agent's unauthorised access to a government portal as two lawmakers move to ban superintelligence
- AI Compliance Map — Sep 11–18, 2026: EU proposes AI companions off by default for minors as a judge finds Montana's deepfake-ad law likely unconstitutional
- AI Compliance Map — Sep 4–11, 2026: California enacts audit and child-safety frameworks as China sets AI dispute rules
- AI Compliance Map — Aug 28–Sep 4, 2026: California reaches the Governor as Ireland puts GenAI court controls into force
- AI Compliance Map — Aug 14–21, 2026: China data-risk rules take effect as US, UK and Australia sharpen AI controls
- AI Compliance Map — Aug 7–14, 2026: Colorado drafts ADMT rules as Texas builds public-sector controls and Australia maps agent risk
- AI Compliance Map — Jul 31–Aug 7, 2026: EU transparency rules go live, while AI-agent access and child-safety scrutiny sharpen
- AI Compliance Map — Jul 24–31, 2026: EU timeline reset, Munich copyright ruling, and AI-pricing antitrust risk