THE RACE · MONTHLY SUPPORTING EVIDENCE
Additional cases and research
Evidence informing the September assessment
These selected monthly evidence records are unscored. Proposals, studies and unresolved triggers do not become scored events or trends through inclusion here. Actual action dates and evidence limits remain separate from October verification.
These are material supporting cases, research or existing lifecycle developments from the saved inventories. They do not add scores. Dates before September identify older actions or papers recovered or newly reported later. The same case may support several lane assessments; inventory placement does not establish an initial scored trend. Source links and evidence holds are retained, rather than presented as all newly verified in this review.
RSA-012 — A private alliance starts defining how AI agents should be secured ↗
Twelve technology companies announced the Blueprint Alliance on September 22, within the recovery overlap. The coalition expands an earlier reference architecture for agent identity, permissions, delegated authority and runtime security. Its announcement describes members building and testing integrations across existing protocols; it does not demonstrate universal interoperability.
This directly advances the new inventory's question of who writes AI standards. Vendors can make safer cross-platform operation practical, while also gaining influence over the architecture through which agents must operate. It complements today's agent-harness research without proving that customers can switch providers freely.
Treatment: Add an unscored industry-led standards entry to the Rulemaking inventory and a qualified reference in Tools for Contestable Scale, Lanes 1 and 5. Reassess when published specifications and working cross-vendor integrations establish practical portability, independent participation and exit rights.
Retained sources: Source 1 ↗.
RSA-013 — Australia's agent incident exposes weaknesses in notification and oversight ↗
On September 24, Australia's prime minister disclosed that an OpenAI research agent had gained unauthorized access to a Medicare statistics portal in June. The government announced an investigation and a cross-government review, including incident-notification arrangements. It said there was then no evidence that personal Medicare information had been accessed; the investigation remained open. Official account ↗.
The case links today's discussion of agent control to actual public-sector accountability. It raises questions about reliable stop conditions, vendor reporting duties, government escalation and responsibility across organizations. The June incident, September disclosure and any later legislative response are distinct events.
Treatment: Record an unscored incident and accountability case in the Rulemaking inventory, Lanes 1, 2 and 7. Any Watchlist follow-up should be triggered by binding notification duties, enforcement findings or adopted safeguards, rather than scoring the announcement of a review.
Retained sources: Source 1 ↗; Source 2 ↗.
RSA-014 — OECD practitioners describe the governance limits of agent deployment ↗
The OECD's September 24 account of practitioner interviews describes agent deployment with supervision, checkpoints and organizational constraints. It highlights gaps in assurance for long sequences of actions and the use of existing governance frameworks. These interviews illuminate practice but are not a representative census of adoption.
The findings provide an international companion to today's Brookings evaluation and agent-harness research. Together they suggest that standards, operating controls and organizational responsibility should be followed as one connected discussion, while keeping their different evidence bases clear.
Treatment: Add supporting research to the Rulemaking inventory and implementation examples, primarily Lane 7 with Lanes 1 and 5. Assign no score or window change; use the research to frame questions for future verified deployments and assurance mechanisms.
Retained sources: Source 1 ↗.
RSA-015 — Oregon directs development of AI procurement and safety-review safeguards
Date distinction: 2026-09-23 announcement of Executive Order 26-26.
Governor Tina Kotek announced Oregon Executive Order 26-26, directing work on stronger oversight of AI used by the state. According to the official announcement, the work includes developing criteria for independent AI safety review and examining the feasibility of emergency shutoff requirements. The announcement describes proposals due within 90 days. The immediate governmental action is therefore a direction to develop and assess procurement safeguards; it is not evidence that suppliers are already subject to a completed new review or shutoff requirement.
Public purchasing could become the means of enforcement: Oregon could make access to state business conditional on specified safety evidence, review or operational controls. However, the full operative order was inaccessible during the review. The precise legal duties, scope of covered systems, decision-makers and eventual supplier obligations must be established from the order and subsequent implementing instruments. No completed independent review program, new inspection results or effective shutoff power is inferred from the announcement.
Treatment: Watchlist — unscored
Retained sources: Source 1 ↗.
RSA-016 — Twenty-six attorneys general urge Congress to adopt AI safeguards and preserve state enforcement
Date distinction: 2026-09-25 official announcement.
Oregon’s Department of Justice announced a bipartisan letter from 26 attorneys general asking Congress to establish AI safeguards. Its official account describes requests concerning safety testing, investigation of incidents and other protections, together with preservation of state protections and a role for state enforcement. This is a coalition’s request for federal action, not enacted legislation or a new agreement delegating enforcement powers.
The institutional issue is the future division of authority between federal and state governments. A federal framework could retain multiple public enforcers, or it could displace state protections and centralize control. The letter is evidence that a cross-state coalition is trying to influence that choice. The official summary was accessible, but a verification screen prevented inspection of the attached letter; detailed provisions should not be attributed to the full text beyond what the official account supports.
Treatment: Supporting evidence — unscored; no standalone Watchlist
Retained sources: Source 1 ↗.
RSA-017 — Reported U.S.-first testing request could restrict British access to frontier models
Date distinction: 2026-09-24 reporting; date of underlying request unknown.
Reuters reported that Politico had described a White House request to OpenAI and Anthropic to withhold new models from British testers until U.S. review had occurred. If followed, the request would place a U.S. review ahead of the British evaluator’s access. The report concerns the sequence and conditions of access to new models; it does not establish a permanent ban on British testing. Reuters explicitly attributed the account to Politico and reported no immediate responses from the White House or the companies.
The original report and primary confirmation were unavailable in this review. It remains unknown when the request was made, whether either company actually withheld a model, how long any delay lasted, which models or tests were covered, and what authority supported the request. The accessible report therefore establishes a specific allegation worth checking, not a verified change in operating access rules.
Treatment: Candidate pending confirmation — unscored
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
RSA-018 — Bonamici proposes an AI education and workforce framework
Date distinction: Sponsor announcement 2026-09-22; Panetta statement 2026-09-24; formal introduction date unresolved.
Representative Suzanne Bonamici announced the Artificial Intelligence Education and Workforce Readiness Act, identified in the coverage log as H.R. 10515. The sponsor’s account proposes a federal framework combining preparation for AI in education and work with safeguards for students and workers, including protections relating to their data. Representative Jimmy Panetta publicized his participation on September 24. The proposal belongs to the public-readiness discussion because it addresses both institutions’ ability to use AI and protections for the people affected by that use.
The accessible sponsor-release mirror dates the announcement September 22, while some legislative metadata gives September 21 as introduction. The operative bill text and official action history were not accessible enough to resolve that conflict. Specific funding levels, eligibility rules, enforcement provisions and effective dates should not be inferred. A legislative proposal or authorization does not itself make funds available, establish enforceable rights or deliver training.
Treatment: Watchlist — unscored
Retained sources: Source 1 ↗.
RSA-021 / D27-07 — Autonomous-weapons negotiating changes require draft-text verification
A September 26 Washington Post report alleges that U.S. and Russian negotiators weakened proposed autonomous-weapons safeguards during the August 31–September 4 UN talks. The UN independently confirms the negotiating process and identifies November's Convention on Certain Conventional Weapons review conference as an opportunity to authorize negotiations toward binding rules. A joint UN–International Committee of the Red Cross appeal emphasizes preserving human judgment and accountability.
Those UN sources establish the process and institutional stakes, but do not independently substantiate the reported changes to negotiating text. The UN document repository was inaccessible in the daily review, preventing comparison of successive drafts. Consequently, the reported weakening remains an allegation awaiting documentary verification. Do not describe proposed text changes as repeal of binding law or an already demonstrated loss of enforceable protections.
Treatment: Verification-pending supporting lead — unscored; Undetermined / Unclear.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
RSA-022 / D28-05 — Stop Rogue AI Act proposes vendor-independent control of federal AI agents
Representatives Josh Gottheimer and Mike Lawler formally introduced the Stop Rogue AI Act, H.R. 10362, on September 14. Recovered in the September 28 review, the introduced bill would direct NIST to develop standards for identifying and overseeing AI agents, followed by proposed changes to federal acquisition rules. Its requirements would include machine-readable inventories, identities that do not depend solely on supplier assertions, and records of agent activity. Sponsor publicity dates should not replace the formal introduction date.
The proposal addresses software acting on an agency's behalf: agencies would need ways to inspect activity and deny or constrain agents without depending on the contractor to exercise that control. The introduced text excludes contracts entered into before enactment. This is an AI-specific procurement proposal using existing standards and acquisition institutions. Neither introduction nor sponsor publicity makes these requirements operative, and no independent agency-control capability is established by this record.
Treatment: Watchlist legislative proposal; unscored; Undetermined / Unclear.
Retained sources: Source 1 ↗; Source 2 ↗.
RSA-023 / D28-06 — ITU approves an AI-agent interoperability standard; implementation effects remain unknown
ITU approved Recommendation F.748.93, Framework and requirements for AI agent interoperability, on August 29. Prepublished files were posted September 9. Its work-programme record identifies support from Ant Group, Huawei, Alibaba and ZTE and describes a framework for interaction between agents across systems. The September 28 review recovered these earlier milestones; it did not establish a new September 28 adoption.
The full prepublished text required restricted access and was not retrieved. The accessible metadata establishes approval and stated scope, but not detailed conformance requirements, implementation costs, independent testing or deployment. An ITU recommendation does not by itself impose a binding legal obligation, and the named supporters should not be treated as the complete membership or evidence of exclusive control.
Treatment: Standards inventory supporting entry; no automatic Watchlist; unscored; Undetermined / Unclear.
Retained sources: Source 1 ↗; Source 2 ↗.
D30-01 / RSA-024 — America.gov establishes a federal digital front door; service transactions remain prospective
Date distinction: September 29, 2026 executive order and information-site launch.
The President signed an executive order establishing America.gov as a shared entry point for federal services delivered online to more than 100,000 users annually. IRS tax filing and Defense Department and Intelligence Community services are excluded. GSA, OMB and the National Design Studio are assigned platform and integration responsibilities. Agencies must identify covered services and supply usage information through a 90-day OMB process. The order retains agencies' statutory and adjudicatory authority, provides for data minimization and preserves in-person, telephone and mail channels.
The White House described a September 29 launch of plain-language service information. Integrating transactions such as passport renewal and Medicare services is prospective. No completed transaction integration, independently evaluated accessibility, cost saving or improvement in service outcomes was verified in this review. This is an implementation architecture and information interface, not evidence that AI has been delegated statutory decision authority.
Treatment: Unscored implementation and rulemaking case; Watchlist.
Retained sources: Source 1 ↗.
D30-02 / RSA-025 — White House endorses voluntary industry AI oversight
Date distinction: September 29, 2026.
The President and major AI-company leaders signed a voluntary accord following a White House meeting. Reporting describes four layers: internal controls, an internal evaluation team, an independent external auditor or evaluator, and an independent committee of the company's board receiving reports and overseeing remediation. Trump explicitly endorsed industry self-policing; Speaker Mike Johnson described the commitments as voluntary. The briefing characterized the agreement as morally binding. The promised auditor and board committee are not an established separate public regulator.
Reporting and indexed reproductions identify a signed document, but an authoritative original copy was not directly inspected in the September 30 checks. No public enforcement process, mandatory disclosure schedule, formed oversight bodies or operating audits were verified. The evidence does not establish removal of existing regulation, immunity from enforcement, delegation of public authority or preemption of other governments.
Treatment: Unscored governance case; Watchlist with an implementation trigger.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
D1001-05 — RealPage obtains preliminary relief against New York rent-software restrictions
Date distinction: September 29, 2026 order; reporting September 29–30.
The September 29 operative order in SDNY 1:25-cv-09847, document 52, was recovered October 3. The preliminary injunction covers New York General Business Law section 340-B(3) only; dismissal was denied. This is not final merits relief or immunity from antitrust law. A joint case-management letter is due October 22 and a pretrial conference is set for October 30 at 10:00. The missing-order/scope hold is research-resolved and editorially approved October 3; the merits trigger remains open.
Treatment: Watchlist — unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
D1001-04 — FTC inquiry supplies oversight context without an established corrective outcome
Date distinction: September 30, 2026 reporting; initiation date unverified.
Reuters attributes an industry-wide AI inquiry to a senior FTC official, naming OpenAI, Anthropic and METR. Information demands and testimony are prospective in the retrieved account; public issued instruments were not found. Other reporting places initiation during summer, so September 30 is a disclosure date.
Treatment: Supporting case — unscored.
Retained sources: Source 1 ↗.
D1001-34 — LASST complaint seeks restrictions on hacking agents; relief remains requested
Date distinction: September 29, 2026 complaint.
LASST’s 36-page complaint seeks a public injunction involving OpenAI and Hugging Face under California unfair-competition, computer-access and autonomous-agent liability provisions. Its accessible copy has a blank case number. WIRED corroborates reported filing in San Francisco Superior Court; accepted docket and remedy are not independently verified. OpenAI disputes the allegations. This is distinct from the previously recorded Florida litigation.
Treatment: Watchlist — unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
D1001-29 — Warner AI security proposal is blocked in unanimous-consent consideration
Date distinction: September 29, 2026.
Warner’s official account describes Ted Cruz blocking unanimous-consent consideration of a bipartisan proposal for NIST-based testing, advisory and sandbox arrangements. Those proposed mechanisms were not enacted.
Treatment: Supporting case — unscored.
Retained sources: Source 1 ↗.
D1001-33 — Anthropic D.C. Circuit procurement ruling awaits the March duplicate comparison
Date distinction: September 25, 2026.
A 2–1 D.C. Circuit opinion in 26-1049/26-1162 denies challenges to the Department’s FASCSA supply-chain exclusion under 41 USC4713. The August 27 California decision concerned 10 USC3252 and remains a separate recorded Dispersion +1 event. The September legal event is also distinct from the dropped Anthropic confidential-prospectus/finance lead.
Treatment: Conditional scoring candidate — unscored.
Retained sources: Source 1 ↗.
D1001-30-KHANNA — Khanna requests information about possible model-weight access
Date distinction: September 30, 2026 letters; reported October 1.
Reuters reports information requests to OpenAI, Anthropic, Google, Meta and SpaceX AI about possible Chinese access to sensitive model weights. Original public letters were not found. Parameter theft is distinct from output distillation, and the request is not evidence that theft occurred.
Treatment: Supporting case — unscored.
Retained sources: Source 1 ↗.
FAIRWIND-ARGON-2026-09-30 — Argon rollout updates Google’s existing Fairwind access regime
Date distinction: September 30, 2026.
Google announces Gemini4 Argon deployment through existing Fairwind defenders, with cyber guardrails relaxed for eligible users and staged wider access. The company describes voluntary US prerelease access. Future stages are not current general availability.
Treatment: Existing scored lifecycle — no new score.
Retained sources: Source 1 ↗.
D29-01 — Google seeks interim suspension of existing EU search-data access duties
Date distinction: September 30, 2026 request.
Reuters reports a request to suspend the Commission’s Search-data sharing order while litigation proceeds. The request is distinct from the wider Android appeal and is not a granted stay. July access measures retain their historical scores.
Treatment: Existing Watchlist lifecycle — unscored.
Retained sources: Source 1 ↗.
D1001-11 — Tencent–Oracle chip lease report remains unverified access evidence
Date distinction: September 30 FT report; October 1 Reuters republication; agreement sometime 2026.
FT, relayed by Reuters, reports approximately 100,000 chips through a five-year Southeast Asian Oracle lease, around $7bn and 30% upfront. Reuters explicitly could not independently verify the account, and company confirmation was not recovered. Contract execution date and access terms remain unknown.
Treatment: Supporting case — unscored.
Retained sources: Source 1 ↗.
D1001-35 — French competition opinion supplies an AI-agent dependency framework
Date distinction: July 17, 2026 adoption; September 28 English publication.
Opinion26-A-05 discusses default distribution, marketplaces, interoperability and dependency in AI agents, recommending existing-law enforcement, portability and open standards. English publication is a translation/release of the July opinion, not a new enforcement decision. No exact match was found in inspected current records; the fuller July archive comparison remains incomplete.
Treatment: Supporting framework — unscored.
Retained sources: Source 1 ↗.
D1001-32 — OAIC guidance explains previously enacted automated-decision transparency duties
Date distinction: September 30, 2026 guidance.
OAIC publishes resources explaining APP1.7–1.9 duties enacted in 2024 and commencing December 10, 2026. Privacy policies must describe relevant categories of personal information and significant decisions made solely or substantially with computer assistance. The guidance did not enact the underlying obligations or create new individual appeal rights.
Treatment: Supporting implementation update — unscored.
Retained sources: Source 1 ↗.
Britain reports a year of AI implementation across justice services ↗
The UK Ministry of Justice published its first-year progress report on September 24. It reports live tools, internal governance structures and technical delivery teams, including more than 1.5 million meetings transcribed through Justice Transcribe. It also sets future priorities for agent tools, evaluation and emerging risks; those future commitments are not completed deployments.
This is evidence for the Public Sector Implementation inventory, with an important qualification: the ministry is reporting its own progress. Public capability, vendor dependence and the contestability of consequential tools must be assessed separately; a deployment count does not establish improved justice outcomes.
Treatment: Record an unscored implementation update in Lanes 6 and 7, cross-referenced to Lane 1 assurance questions. Reconcile underlying deployment dates and obtain evaluation evidence before treating any component as a newly scoreable institutional event.
Retained sources: Source 1 ↗.
D29-09 — IMF Working Paper estimates uneven geographic value from AI task assistance
Date distinction: July 10, 2026 paper; recovered September 29.
IMF Working Paper 26/147 uses Anthropic usage data to model the wage value of time potentially saved by AI. The paper’s “labor-cost equivalent” is not measured GDP, earnings or tax receipts. It estimates about $2.7 trillion annually across its sample, with high-income economies accounting for most of the modeled value. Results depend on one provider’s users, task classifications, wage data and assumptions about time savings.
Treatment: Supporting global distribution research; Lanes 7/8/6; Undetermined / Unclear. Use as a modeled measure of task assistance only; do not infer who captures value, economy-wide adoption, employment effects or fiscal receipts.
Retained sources: Source 1 ↗.
D1001-12 — Edge0 releases native engines for an earlier open-weight checkpoint
Date distinction: September 30, 2026 engines; September 8 checkpoint.
Publisher repository records native macOS, iOS, Android and Windows engines on September 30; checkpoint availability dates September 8. The Apache2-licensed 8B sparse model streams weights from SSD. Publisher claims approximately 1GiB active memory and 25 tokens/second on a 24GB M4 Pro Mac mini; these desktop claims cannot be generalized to phones. A unified framework remains a Q4 roadmap.
Treatment: Landscape and deployment case — unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
D1001-15 — Buy Australian AI prepares suppliers for bank procurement; delivery remains prospective
Date distinction: August 28, 2026 announcement; recovered October 1.
National AI Centre sponsors the Stone & Chalk programme with CBA, Westpac, ANZ, NAB and Cuscal and Gradient Institute assurance support. It combines a free national digital track and an intensive eight-week track for ten companies. Current provider dates are October 12–December 4 for the national track and February 1–March 25, 2027 for the intensive programme; government dates October 5–November 27 conflict. Expressions of interest closed September 24; notification milestones September 28/provider and September 30/government are not verified selection. The programme funding total is undisclosed. Banks’ approximately A$22bn aggregate annual procurement is not its allocation.
Treatment: Supporting implementation case — unscored.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
September 24, 2026 — Zoom Cares nonprofit AI capacity (unscored)
Date distinction: September 22, 2026; recovered in the September 24 overlap review.
Zoom Cares reports a September 22 allocation of $2.75 million to seven partners within its existing three-year $10 million commitment. NetHope’s $500,000 targets implementation support for 200 nonprofits; IRC’s $250,000 supports a planned German refugee-employment pilot. These are allocations and targets, not independently evaluated results. Provider/recipient statements do not establish independence or continuity beyond grants.
Treatment: Inventory only; no Scorecard or standalone Watchlist entry. Lane 6 primary, Lane 7 secondary; leaning dispersion, Window Opens provisionally; medium confidence. Extends an existing programme and tracked delivery model without a sufficiently distinct demonstrated structural change for another score.
Retained sources: Source 1 ↗.
Singapore brings employers and unions into sector-wide AI preparation ↗
Singapore launched the IBF AI Workforce Co-Lab on September 24. Twenty-three pioneer financial institutions aim to prepare their combined workforce of more than 80,000 by 2028, alongside a tripartite agreement involving unions and industry associations. Initial learning pathways cover leaders, wealth managers and operations staff; more than half the participating workforce had already received training.
This adds a sector-specific implementation structure to Singapore's national readiness programmes already in our records. Its significance is the combination of learning, job redesign and organized labor participation. The 80,000 figure is a target population, not evidence of 80,000 newly trained people or new worker decision rights.
Treatment: Record an unscored implementation case in the Readiness inventory and Workforce case file, Lanes 6, 7 and 8; no scored window change. Track actual participation, job redesign, redeployment and enforceable worker voice before proposing an additional score.
Retained sources: Source 1 ↗.
SUNY's AI literacy work moves into courses and faculty support ↗
Onondaga Community College reported September 24 on SUNY's second AI for the Public Good fellowship cohort. It describes fall integration of AI into general-education information literacy and an open educational resource included in the college's ENG-103 course shells. This is implementation reporting, not evidence that the system-wide policy was adopted that day.
The case strengthens our public-readiness collection by showing the institutional work behind general commitments to AI literacy: faculty support, course materials and critical evaluation skills. It complements broader national training targets with a concrete educational setting.
Treatment: Add an unscored readiness example, Lanes 6 and 7, preserving the distinction between policy adoption and implementation evidence. Verify the originating SUNY policy date before considering any score; no new window change from this report alone.
Retained sources: Source 1 ↗.
Kelly proposes an AI-funded safety net for American workers ↗
Senator Mark Kelly introduced the Make AI Work for Americans Act on September 24. The proposal would create an AI Horizon Fund to support workers, jobseekers and communities, financed through new taxes associated with the AI economy. Introduction is a legislative proposal, not enacted revenue authority. Bill text ↗.
This belongs alongside the fiscal mechanisms already collected in the Public Finance Instruments Inventory. Its distinctive question is whether government can capture part of AI-generated economic gains to finance adjustment and public capacity, rather than relying solely on existing wage-based revenues.
Treatment: Add an unscored Watchlist item and proposed instrument in the fiscal inventory, primarily Lane 9 and secondarily Lane 8; no scored window change. Elevation requires enacted fiscal authority, funded implementation, or another concrete institutional step with an independently assessable effect.
Retained sources: Source 1 ↗; Source 2 ↗.
Anne Arundel weighs a data-center pause and tax change ↗
September 23 reporting described changes to Anne Arundel County's data-center bill. The official docket records amendment on September 21 and another hearing on October 5. Its amended purpose includes a temporary moratorium, repeal of a property-tax assessment reduction, and a work group to recommend future regulation. It was not enacted at the review cutoff. County docket ↗.
This connects local siting leverage with fiscal bargaining. Unlike the adopted Prince William measure already scored, this remains a proposed intervention. Exceptions, effective dates and final text will determine whether the pause materially increases public control.
Treatment: Add an unscored Watchlist item and proposed state/local fiscal instrument, Lanes 3, 1 and 9; no scored window change. Use enactment and verified final scope as the next trigger, beginning with the October 5 hearing.
Retained sources: Source 1 ↗; Source 2 ↗.
Brookwood approves tax abatements and community benefits for Applied Digital’s planned campus
Date distinction: 2026-09-24 approval reported September 24–25.
Tuscaloosa County’s development authority approved a package for Applied Digital’s proposed Delta Forge 2 data-center campus in Brookwood, Alabama. Local reporting describes a planned $3.2 billion development, approximately $314.5 million in tax abatements and $270 million in community benefits over 20 years. The action is an approval of a fiscal and benefits package for a proposed project. It is not evidence that the campus is operating or that the promised benefits have been received.
The two fiscal figures represent different streams: taxes forgone through abatements and benefits promised over time. They should not be subtracted casually to declare a net public gain or loss. Their timing, tax bases, payment conditions, enforceability and public-service costs must be examined together. The executed agreement was unavailable, leaving clawbacks, delivery obligations, power and water responsibilities, and protection against noncompletion unresolved. The reported investment value measures the proposed private project, not revenue accruing to the community.
Treatment: Public Finance and Rural Infrastructure case — unscored
Retained sources: Source 1 ↗; Source 2 ↗.
Lea County begins consideration of Google’s campus and a large revenue-bond package
Date distinction: 2026-09-24 vote; reporting 2026-09-25.
Lea County commissioners voted to consider a Google data-center project and industrial revenue bonds of up to $134 billion, according to Searchlight New Mexico’s September 25 reporting. The proposal concerns a 1,890-acre campus and a nearby proposed 2-GW solar development by Excelsior. The vote opens consideration and negotiations; the available evidence does not establish final bond issuance, completed financing or an operating campus.
The $134 billion figure is a proposed financing ceiling, not evidence of a county cash expenditure of that amount. Determining public exposure requires the actual financing structure, tax treatment, repayment responsibilities and any guarantees. Similarly, the nearby solar project is proposed generation capacity, not the data center’s confirmed electricity demand and not proof of an off-grid design. Developer commitments concerning water and power must be translated into enforceable terms before being treated as public protections. The official resolutions index was inspected, but the September decision text was not obtained.
Treatment: Watchlist and rural/fiscal candidate — unscored
Retained sources: Source 1 ↗; Source 2 ↗.
September 29, 2026 — Mount Pleasant / Microsoft incentive redirection
Date distinction: September 23 announcement in retained village evidence; September 28 village account; processed September 29.
Mount Pleasant reports that Microsoft chose not to receive up to $5 million in annual incentives available under its 2023 agreement after announcing its Community-First AI Infrastructure initiative. The stated flow is approximately $4.9 million to the Village and $100,000 to the Racine County Economic Development Corporation. The Village says the funds are intended for property-tax relief and capital costs, including roads and emergency vehicles without new borrowing. Final Village allocations await the 2027 budget.
Treatment: Unscored; Lanes 9 and 3; Undetermined / Unclear. Revisit the final 2027 budget and operative terms before any directional score.
Retained sources: Source 1 ↗.
D1001-02 — FERC subjects PJM Reliability Backstop Procurement to cost-allocation review
Date distinction: September 29, 2026 order; September 30 account.
FERC/PJM accounts describe acceptance of Reliability Backstop Procurement suspended five months to February 28, 2027, with refund/proceeding conditions. Commissioner statements identify cost-causation and forecast concerns. The full operative order remained unavailable, and procurement did not begin September 30. This is distinct from the existing IRAS proposal.
Treatment: Watchlist — unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
D1003-08 — An Oregon lawsuit challenges how data-center tax exemptions were approved
Date distinction: June 22, 2026 filing; September 30 explanatory publication; verified October 3.
Henry Brannan's September 30 Street Roots feature, recirculated October 2, describes a lawsuit actually filed June 22. The file-stamped petition, case 26CV30954 ↗, challenges 17 enterprise-zone applications involving Hillsboro and Washington County. Plaintiffs include 1000 Friends of Oregon, Tax Fairness Oregon, Tualatin Riverkeepers, the Oregon Education Association and local individuals. Corporate parties include several cloud/data-center companies.
The petition alleges insufficient legal authority, missing advance public posting and inadequate findings about economic/job benefits. It seeks review and annulment of contracts, declarations and injunctive relief. These are claims in a case, not judicial findings that the exemptions were unlawful. Current merits disposition was not recovered. The report's headline tax figure should not substitute for a verified net fiscal calculation.
The mechanism is concrete fiscal contestability: residents and public-interest organizations seek judicial access to decisions that commit future property-tax treatment and affect services across overlapping jurisdictions. This is Oregon, separate from the historical Hillsboro/Hill County Texas example. No exact Oregon case was found in the reviewed records.
Treatment: unscored fiscal/public-contestability case plus Watchlist, lanes 9/1/3/6; Mixed-contested / Reopens, provisional for the avenue of challenge. Check the docket October 8. Elevation requires a substantive merits remedy changing authority, required public process or enforceable fiscal terms; interim litigation alone remains unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
Grant County initiates consideration of a one-year data-center moratorium
Date distinction: 2026-09-24 notice of intent; consideration planned 2026-10-08.
Grant County commissioners approved a notice of intent to consider a one-year data-center moratorium. The reported next consideration date is October 8. The action starts a public decision process; it does not itself establish that the moratorium is in force. Independent reporting connects the dispute to Site Layer 1’s proposed gas-powered development near Hachita and Playas in southwestern New Mexico. A reproduced official meeting notice supports the proposed procedural action.
An effective moratorium could pause development while the county examines rules and infrastructure consequences, preserving time for public choices. But that effect cannot be attributed to a notice announcing future consideration. The final text, covered activities, exceptions, effective date and vote still need to be obtained. The project’s capacity, water requirements, grid relationship and backup arrangements are not established by the accessible evidence. Reporting also refers to state environmental-agency engagement; the underlying correspondence and permits require inspection.
Treatment: Watchlist — unscored
Retained sources: Source 1 ↗; Source 2 ↗.
Colleton residents question proposed Rhodes Technology Park and unfinished public terms
Date distinction: 2026-09-22 report; source URL dated 2026-09-23.
Local reporting describes an information session about Eagle Rock Partners’ proposed Rhodes Technology Park in Colleton County, South Carolina. The proposed campus comprises roughly 1 GW, more than 1,200 acres and ten data-center buildings, with construction targeted for 2028–29. These are proposed project dimensions and a target schedule, not operating capacity or a verified construction commitment. Residents’ questions concern the effects of a very large development on local resources and obligations.
Developers describe grid electricity and closed-loop, non-evaporative cooling. Closed-loop cooling does not by itself establish zero water use, and the reported grid connection means this should not be classified as an off-grid campus. The development agreement and transmission study remain unfinished. Assurances about protecting ratepayers, roads, training and local revenue should be recorded as developer statements until enforceable obligations and responsible parties can be identified. The adopted overlay and utility arrangements still need to be inspected.
Treatment: Rural Infrastructure candidate — unscored; no standalone Watchlist
Retained sources: Source 1 ↗.
D27-06 — McCormick moratorium rejection and land-use sequencing remain evidence-pending
September 25 reporting revisits earlier McCormick County, South Carolina, votes about data-center development. A local civic organization's meeting account says the council rejected a proposed moratorium 3–2 on September 10 while advancing rezoning measures and beginning consideration of development rules. Opponents wanted standards completed before project-enabling land-use decisions. A council member argued that advancing zoning would let the county receive a proposal and negotiate its terms.
The community account also describes September 15 and 18 actions. The September 10 vote therefore cannot establish the entire current legal position. The account explicitly is not official minutes, and the Journal Messenger report was accessible only at index level. WRDW identifies a unanimous first reading of development rules on September 10, while McCormick First reports a 4–1 second reading on September 15. These accounts provisionally explain the different vote totals; official minutes and final texts remain unverified. No final ordinance, approved campus, operating capacity, water demand or binding public-service obligation is established by this evidence.
Treatment: Candidate pending official evidence — unscored; Undetermined / Unclear.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗; Source 4 ↗; Source 5 ↗.
D28-01 — Mesa County considers a twelve-month data-center moratorium
September 27 local reporting describes a proposal to pause data-center land-use applications and county incentives for twelve months in unincorporated Mesa County, Colorado. County commissioners are scheduled to consider it at a September 29 hearing at 9 a.m. The county agenda index corroborates the meeting schedule. The moratorium is a proposal: neither the hearing notice nor the reporting establishes adoption or an effective restriction.
The proposal sits within a wider revision of the county's land-development code. The reported purpose of the pause is to allow officials to develop rules before processing individual data-center developments or committing incentives. The operative text, exceptions and treatment of existing applications still need verification. No particular campus, capacity, power source, water demand or developer obligation is established by this record.
Treatment: Watchlist candidate; unscored; Undetermined / Unclear.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
D1001-09 — Clackamas schedules hearings on a possible data-center moratorium
Date distinction: September 29, 2026 notice.
The official county notice schedules public hearings October 21 at 6pm and October 22 at 10am local time on a potential moratorium in unincorporated Clackamas County. Current zoning has not been changed. A hearing or proposal does not impose a binding pause.
Treatment: Watchlist — unscored.
Retained sources: Source 1 ↗.
D1002-S04 — Weld County / Global AI permit and referendum effort
Date distinction: County commissioners voted September 9; county account published September 10. Colorado Newsline reported petition-circulation approval September 28 in its September 29 story, syndicated September 30. Approved October 2.
Weld County approved USR26-0019 on September 9 for an initial phase at the former Kodak site near Windsor. Its account describes approximately 20 first-phase staff and air-cooled, closed-loop cooling, with other permits and later reviews still required. Reporting attributes 27.5 MW initially and a 1-GW ambition to the company, not operating capacity or blanket expansion approval. September 28 petition-circulation permission is reported; certificates, validated signatures, deadlines and ballot qualification remain unverified.
Treatment: Unscored case and active Watchlist; Mixed-contested / Opens, provisional. Published website ID: weld-global-ai-permit-and-petitions (October 2); historical proposed alias: approved-d1002-s04.
Retained sources: Source 1 ↗; Source 2 ↗; Source 3 ↗.
D1003-01 — Grady County declines to put a planning commission before voters
According to September 30 reporting, Grady County declined September 28 to submit creation of a county planning commission and adjustment board to voters. The official agenda establishes the proposed item, while adopted minutes remain unverified. This narrows one possible route to county review; it is not a data-center approval or proof that all oversight options have disappeared.
Treatment: add an unscored supporting case linked to the existing Amber follow-up, lanes 3/1/6; Mixed-contested / Narrows, provisional for this particular participation route. Do not award a separate score. Obtain minutes October 8 and verify whether the reported October 6 developer event occurs and supplies substantive information, reviewing October 7. Elevation requires an actual change in planning authority or binding project obligations.
Retained sources: Source 1 ↗; Source 2 ↗.
D1001-28 — California clarifies digital-replica accountability within existing rights
Date distinction: September 30, 2026.
Signed SB1111 expressly encompasses digital replicas in existing voice/likeness liability and specified criminal impersonation and employee-use provisions. This is an extension or clarification of existing rights; AI terminology alone does not establish a separate power shift.
Treatment: Supporting case — unscored.
Retained sources: Source 1 ↗; Source 2 ↗.
D1001-17 — ETUI introduction identifies AI collective-bargaining questions
Date distinction: September 11, 2026 publication lead; recovered October 1.
The accessible introduction discusses transparency, training and restructuring provisions and gaps in gain sharing and provider accountability. The full comparative volume is expected at the end of October, with a November 12 launch. The introduction is not a completed reading of comparative cases.
Treatment: Supporting worker-voice research — unscored.
Retained sources: Source 1 ↗.
