THE RACE · MONTHLY SUPPORTING EVIDENCE
Political influence and public capacity
Provisional · Evidence through October 31, 8:30 a.m. MDT · Historical source checks November 1
Political influence and public capacity
Competing AI policy networks are deploying political resources, but competition among well-funded factions does not establish dispersed agenda control. The mechanism to track is conversion of money, access and organizational continuity into public rules. Election activity alone receives no score here.
| Network or entity | Verified evidence and date boundary | Interpretation and limits |
|---|---|---|
| Public First Action and affiliates | Anthropic’s July 21 statement says its total support reached $40 million and that its donations cannot fund candidate-election influence. Public First Action’s own news page retains the June $80 million fundraising claim and July $15 million project announcement. The separately registered Public First committee’s FEC summary covers through September 30: receipts $4.91 million; $4.34 million contributed to other committees. | Corporate advocacy, a 501(c)(4), an affiliated political network and one FEC committee are different entities. Do not label Anthropic’s donations PAC election spending, or infer an $80 million October receipt from an older announcement. Exact affiliate transfers and donor-level reconciliation remain incomplete. |
| Leading the Future and affiliates | FEC C00916114 summary through October 14: receipts $75.98 million; disbursements $52.16 million, including $47 million to other committees; ending cash $23.82 million. | This committee is a funding hub; its own summary reports zero direct independent expenditures, which does not mean affiliates spent nothing. Network commitments, transfers and downstream spending must not be added as separate money. Retain earlier named-donor and pledge history rather than count unfulfilled promises as receipts. |
| Guardrails Alliance / Guardrails Action | FEC C00949487 through October 14 reports receipts $3.25 million and independent expenditures $1.55 million. October 7 reporting describes a planned $1.2 million opposition effort; Guardrails Action is separately described as the coalition letter’s nonprofit organizer. | Announced spending and reported spending have different periods and may overlap. The PAC and nonprofit are not interchangeable; neither figure establishes legislative influence or election results. |
The October 27 Zoned Out study adds a different source of influence evidence: unequal access to UK data-center policymakers, with business meetings dominating the investigated records. Its finding is about documented access, not proof that a meeting caused a decision or every public interaction is known. This is relevant counterevidence to local participation gains.
Assessment: continuing concentrated funding and privileged agenda access, with organized competing positions and some civic coalition-building. Spending alone does not establish a policy outcome. The next test is actual policy effects and whether networks persist as lobbying/policy institutions after the elections. The election remains future at the October cutoff; no outcome is inferred.
Sources checked November 1 for the stated older periods: Anthropic ↗, Public First Action news ↗, Public First FEC ↗, Leading the Future FEC ↗, Guardrails FEC ↗, Guardrails’ own advocacy page ↗, October 7 coalition/spending reporting ↗, and Spotlight report ↗.
State and municipal public AI capacity
The public layer is becoming more complete in resources, governance and practical support. Evidence that it changes major supplier terms or permits low-cost exit remains weaker. Dependencies can deepen at the same time that local competence improves.
| Function | October evidence or continuing institutional base | Remaining test |
|---|---|---|
| Visibility | Public data interfaces, agency inventories, GSA records/disclosure provisions, Pflugerville’s mandated public page | Completeness, usable independent inspection and implementation of the new duties |
| Rules | Judicial approval/human responsibility, federal purchasing terms, adopted municipal technology review | Incorporated contracts, scope/exceptions and actual enforcement |
| Experimentation | Public-benefits awards, city competition, Louisiana court development, continuing state assistants | Evaluated service quality, affected-user participation, safe failure and limits |
| Workforce capacity | Educator funding, conditional training, regional support and county work-specific coaching | Skill transfer, paid learning, retention and authority to act |
| Shared infrastructure | GovAI/MRSC/TXShare, existing USAi/NAIRR, reusable public outputs and local servers | Sustainable operating capacity, neutral access, duplication avoided and retained learning |
| Contestability | Elected oversight and review, professional/worker obligations and available model procurement resources | Effective challenge, audit access, remedy and material changes in vendor behavior |
| Portability/independence | GSA terms, local/open components, common application interfaces, provider-choice funding | A working transfer of data, permissions, workflows and institutional knowledge |
| Jurisdiction or network | Current assessment, with dates and stage kept distinct |
|---|---|
| California | September laws and AskCA remain the institutional baseline. October 13 Engaged California publishes participation findings; this gives public priorities visibility, not a binding allocation of authority. September 18 oversight implementation directives remain directives rather than proof that every contemplated audit operates. |
| Maryland | Existing inventory, intake and risk-review architecture continues. Community Compass is a civic platform, not proof of pooled government bargaining. The October public-benefits award includes Maryland/Colorado project work; awards do not establish operational results. |
| Colorado | OIT’s governance/risk/community-of-practice resources remain current background. An October 13 report describes a newly staffed AI architecture role; no independent retention or service outcome follows from the appointment. Douglas County’s October 30 staff account illustrates continuing support and reusable human-reviewed workflows, with scaling limits and self-reported results. |
| Pennsylvania | Official April 15 account describes expansion to more than 3,000 employees and 35 agencies, approved tools, training and a governing board. These are earlier operating-capacity anchors, not new October launches; reported time saved is self-report. |
| Illinois | September 22 EO2026-07 establishes an advisory AI Cabinet and a work programme. Targeted checking did not establish a new completed October procurement or oversight mechanism. That is a bounded result, not proof of inactivity. |
| New Jersey | The July 2024 state assistant and learning programme remain an operating baseline. Targeted checks recovered no separately established October institutional change. State hosting and no-training promises do not automatically establish complete vendor exit. |
| Washington | The July 1 task-force report and earlier policy remain baseline. The October 9 WaTech account describes an October 8 roundtable and the statewide strategy; a meeting is not a new binding safeguard or evaluated capability. |
| San José / GovAI Coalition / MRSC / TXShare | Continuing resource-sharing and procurement capacity supports the existing municipal-network assessment. October meeting and future summit listings show activity, not verified contractual concessions. Preserve the distinction between this coalition and commercial firms using similar “GovAI” branding. |
| NLC, ICMA, CML and county associations | NLC’s toolkit/policy and NACo’s July advocacy/resource hub remain earlier connective capacity. CML’s October issue and ICMA’s October accounts translate practice for local officials. Professional publishing and a sponsored webinar are not themselves new public authority; member-only materials and vendor assertions retain limits. |
Targeted source checks used official state and association sources, including California October 13 ↗, Pennsylvania April 15 ↗, New Jersey’s earlier programme ↗, Illinois order ↗, Washington October account ↗, NACo hub ↗, CML issue ↗, and Douglas County practitioner account ↗. Some checks relied on indexed source text; they do not establish that every underlying instrument was inspected.
