AI Compliance Map — Aug 28–Sep 4, 2026: California reaches the Governor as Ireland puts GenAI court controls into force

The exact coverage window is 28 August 2026 at 17:00 through 4 September 2026 at 17:00, UTC-05:00. Five verified actions met the date and evidence test. California sent a legal-practice AI bill to the Governor, and Ireland brought court-use controls for generative AI into operation. Canada, Shenzhen, and South Korea supplied non-binding infrastructure or policy signals.

The immediate workflow is legal-document control. Teams with California legal-practice exposure should review confidentiality, output verification, court disclosure, and citation procedures. Teams filing in Ireland's High Court or Court of Appeal should apply the new directions now. The other three items belong on infrastructure, product-policy, and standards watchlists.

Action list

Review California legal-service workflows against SB 574. The bill passed the Legislature and went to the Governor on 1 September. Its proposed controls cover delegation of legal work, confidential information, verification of AI output, court disclosure, citations, and arbitration. 1 2
Apply Ireland's GenAI court-document controls to filings made from 1 September. The High Court and Court of Appeal directions require human verification and create disclosure and professional-risk consequences for covered documents. 3
Put data-centre project teams on Canada's responsible-development watchlist. The 3 September principles package asks proponents to account for local benefits, grid costs, water and environmental effects, transparent impact information, and strategic value. The package is a national framework supported by signatories, so the near-term task is project and procurement readiness rather than a new statutory filing. 4 5
Track Shenzhen's 2026–2028 AI plan as a municipal policy signal. Shenzhen's industry, development and reform, and science and technology authorities posted the government-approved plan on 4 September. The plan points toward AI use in health care, finance, employment, education, public administration, courts, data, models, and computing. It creates a policy direction for local engagement and product planning, rather than a new private-sector compliance duty. 6
Keep South Korea's G20 AI standards discussion in the international-policy watchlist. Korea's trade and industry ministry reported on 3 September that its minister discussed AI intellectual-property policy, AI standards, and resilient supply chains at the 2 September G20 Innovation Ministerial Meeting. The release creates a coordination signal for standards and training-data policy, with no direct obligation for companies. 7

Scan table

Date	Jurisdiction and actor	Action and status	Affected scope	Compliance impact
1 Sep	California Legislature; Governor's office next	SB 574 passed the Legislature and was sent to the Governor; pending enactment	Attorneys, courts, arbitrators, parties, and systems handling legal-work information	Prepare controls for confidential-data entry, output verification, court disclosure, citation checking, and limits on delegating legal or arbitral decisions. 1 2
1 Sep	Ireland, High Court and Court of Appeal	Practice directions became operative; court-issued procedural controls	Parties, lawyers, witnesses, experts, and third parties producing civil-proceeding documents	Add independent verification, evidence authentication, directed disclosure, and affidavit, witness-statement, and expert-report controls to Ireland litigation workflows. 3
3 Sep	Canada, Innovation, Science and Economic Development Canada	Responsible Data Centre Development Principles launched; national, non-binding framework	Data-centre proponents and AI infrastructure or cloud projects seeking Canadian sites and approvals	Prepare evidence on grid-cost allocation, water and environmental effects, local and Indigenous engagement, transparency, security, continuity, and access protections. 4 5
4 Sep	Shenzhen municipal authorities	2026–2028 AI Action Plan posted; municipal policy plan	AI research, industry, consumption, public services, governance, and international cooperation in Shenzhen	Treat the plan as a local policy and partnership signal; monitor sector-specific implementation, public procurement, and standards work. 6
3 Sep	South Korea, Ministry of Trade, Industry and Resources	G20 ministerial-meeting report; non-binding international coordination	AI IP policy, standards, and resilient technology supply chains	Track future ISO/IEC and training-data policy work; the release itself changes no company duty. 7

United States: California moves a legal-practice bill to the Governor

California SB 574 passed the Legislature and advanced to the Governor's desk on 1 September. The sponsor's announcement records the legislative milestone. The bill text supplies the proposed controls and remains the governing source for the scope of the measure. 1 2

The proposal would bar an attorney from delegating the practice of law to generative AI. The proposal would also restrict entering confidential, personal-identifying, and other nonpublic information into certain generative-AI systems. Attorneys would take reasonable steps to verify AI output and correct erroneous or hallucinated output. 2

The proposal would require disclosure of generative-AI use to the court for documents submitted to the court. The proposal would also prohibit unverified citations in court filings, including citations supplied by generative AI. Arbitrators would face limits on delegating decision-making to generative AI and on relying on AI-generated information outside the record without disclosure. 2

Compliance impact: legal departments should inventory California-facing workflows that put privileged or confidential material into generative-AI tools, generate court documents, or support arbitration. A review pack should cover approved tools, data-handling rules, human verification, citation checks, disclosure language, and escalation for hallucinated output. SB 574 remains a pending bill until the enactment record changes.

The strict-window review found no qualifying US federal AI action for this edition. The California bill therefore supplies the US legislative entry that can be substantiated within the window.

Europe: Ireland's court directions are operative

The Courts Service of Ireland announced on 31 August that new High Court and Court of Appeal practice directions on responsible generative-AI use in court documents would come into operation on 1 September. The announcement identifies the directions as court-issued procedural instruments, and Ireland's action is a member-state development rather than an EU-level Commission action. 3

The directions cover civil-proceeding documents such as pleadings, submissions, affidavits, witness statements, and expert reports. The covered actors include parties, lawyers, witnesses, experts, and third parties required to produce documents. The directions apply to proceedings instituted before or after commencement and to documents produced after commencement. 3

The required controls include independent human verification, fact-checking, proofreading, confirmation that cited authorities exist and support the proposition, and authentication of exhibits, images, recordings, and other evidence. The directions allow the Court to require disclosure of material generative-AI use. Affidavits and witness statements carry additional declaration requirements, while expert reports carry disclosure requirements for material GenAI use. 8 9

The directions identify possible consequences including adverse costs, rejection or striking out of documents or claims or defences, referrals to the Legal Services Regulatory Authority or a professional body, and personal costs orders. 3

Compliance impact: Ireland litigation teams should add a named human verifier, preserve the provenance of evidence and authorities, record material GenAI use, and route affidavits, witness statements, and expert reports through the required declarations and disclosures. The controls apply to court-document use described by the directions, rather than to every internal use of GenAI.

The strict-window review found no qualifying EU-level AI action and no qualifying UK action with a governing date for inclusion. Ireland's operative court directions provide a verified European entry at the national level.

China: Shenzhen sets a municipal AI policy direction

The Shenzhen Municipal Bureau of Industry and Information Technology, Development and Reform Commission, and Science and Technology Innovation Bureau posted the Shenzhen Artificial Intelligence Action Plan (2026–2028) on 4 September. The notice says the municipal government approved the plan, which covers the period from 2026 through 2028. The underlying notice bears a date of 28 August; this edition uses the portal's 4 September posting as the source-owned event date. 6

The plan calls for broad AI integration across research, industry, consumption, public services, governance, and international cooperation. The operational areas named in the notice include compliant medical-AI applications, finance and fraud or risk identification, employment and education, public-sector assistants and approval processes, AI-assisted court and prosecutorial work, datasets, models and toolchains, computing infrastructure, and international standards. 6

Compliance impact: companies operating in Shenzhen should treat the plan as a local-government policy and industrial-development signal. Product, public-sector sales, infrastructure, and standards teams can use the named sectors to prioritize local engagement and monitor later implementation notices. The plan itself supplies no basis for assigning a new private-sector legal duty in this issue.

The strict-window review found no qualifying China-national AI regulatory action. Shenzhen's plan supplies the China coverage as a clearly labelled municipal policy item.

Other jurisdictions: infrastructure and standards signals

Canada: responsible data-centre development principles

Innovation, Science and Economic Development Canada launched the Responsible Data Centre Development Principles on 3 September. The principles address data centres that power AI and set five expectations: lasting local benefits, no transfer of project-driven electricity costs to Canadians, reduced water and environmental impacts, transparent local-impact information, and strategic value to Canada. 4

The detailed principles ask proponents to pay project-driven grid costs, measure and report water and environmental effects, engage local governments and Indigenous rights holders, provide independently verifiable information, and meet appropriate Canadian legal-protection, cybersecurity, continuity, and access requirements for sensitive-data and critical-infrastructure facilities. The package is a national framework and principles initiative supported by signatories. 5

Compliance impact: AI infrastructure developers and cloud providers should add the principles to site-selection, permitting, utility, public-affairs, and procurement reviews. The immediate work is evidence preparation for proposed projects and local approvals; the principles create no new statutory deadline identified in the release.

South Korea: G20 coordination on AI policy

South Korea's Ministry of Trade, Industry and Resources registered a release on 3 September about the 2 September G20 Innovation Ministerial Meeting. The release says Korea's minister discussed intellectual-property policy for AI, standards for AI, and resilient supply chains. Korea will continue work through ISO/IEC standardization and referred to its fair-use and guideline experience for AI training. 7

Compliance impact: standards, intellectual-property, and government-affairs teams should monitor later national or ISO/IEC work that could turn this coordination signal into technical or policy expectations. The ministerial discussion itself is non-binding.

Near-term deadline register

Effective now — Ireland's High Court and Court of Appeal directions. Apply verification, disclosure, evidence-authentication, and document-specific declaration controls to documents produced after the 1 September commencement. 3
Pending Governor action — California SB 574. The bill reached the Governor's desk on 1 September. The sources reviewed for this edition provide no Governor signature or veto date, so teams should monitor the enactment record rather than assign a date. 1
2026–2028 planning period — Shenzhen AI Action Plan. The municipal plan covers this period; the notice supplies a planning horizon rather than a company compliance deadline. 6
Project-by-project readiness — Canada's principles. The release and detailed principles identify expectations for data-centre development but set no statutory filing date in the material reviewed here. 4

Coverage note

This issue contains five verified in-window actions: California SB 574; Ireland's High Court and Court of Appeal directions; Canada's Responsible Data Centre Development Principles; Shenzhen's 2026–2028 AI Action Plan; and South Korea's G20 policy release. The entries cover the United States, an EU member state, China, Canada, and South Korea.

US federal, EU-level, UK, and China-national qualifying actions remain unverified for this strict window. The review also found no in-window court merits ruling or regulatory-enforcement action that met the source and governing-date test. Those results describe the evidence admitted to this edition; they do not establish that no other activity occurred.

The operational split is narrow: Ireland's court controls are already operative, California's legal-practice controls remain pending enactment, and the Canada, Shenzhen, and South Korea items are non-binding signals. Compliance teams can therefore route the first two to immediate legal-workflow review and the last three to forward-looking infrastructure, product-policy, and standards monitoring.