The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← October 2026 overview

OCTOBER 2026 · LANE 05

Platforms, Distribution, and Separations

Can people change providers, connect alternatives and take their work with them?

Provisional · Evidence through October 31, 8:30 a.m. MDT · Updated November 1, 2026

Direction of power
Mixed-contested
Window
Opens at specific interfaces; Unclear for full exit
Confidence
Moderate
Movement since September
More bounded interfaces and stewardship alternatives; control of context and admission persists

October assessment

October separates more layers without establishing a universal right to leave. SAP’s inbound agent route allows an outside assistant to call SAP agents, while SAP retains permissions and business context. AnyLanguageModel maintains common application code across local and hosted options; the central switching capability predates October, so the release is not a fresh award for inventing it. Foundation stewardship for agent projects may reduce unilateral control of standards, but decision rules and actual transferred rights remain consequential.

GSA’s purchasing provisions add a contractual path for data control and migration in covered federal purchases. This is distinct from a vendor voluntarily shipping an interface. The earlier adoption and October commencement remain one lifecycle. Model choice inside another firm’s workflow is narrower than portable identity, permissions, stored context, action records and operational knowledge.

Agent commerce is adding credential and delegated-payment mechanisms, with incumbents helping define admission and trust. Public protocols can support substitution while creating new dependencies on issuers, permission managers or platforms. Meta’s disputed commercial-content terms and revised November 20 date remain a live counterexample; a delay does not settle the contractual concern.

What would change this judgment

Tested migration of workflows and permissions; binding admission/exit terms; neutral standard-setting and appeals; shipped multi-provider implementations; actual effect of Meta’s final terms after November 20.

Supporting scored events

Each action retains one event and score across lanes. Dates, jurisdictional limits and unresolved verification remain explicit.

October scored action

Google and Constellation combine a major power agreement with a five-year AI relationship

2026-10-06 · Concentration +2 · Secondary lane; primary Lane 3

An October 6 company announcement, published before the cutoff, describes a 20-year power-purchase agreement supporting 890 megawatts of additional nuclear capacity through upgrades to 11 existing reactor units in Illinois, Pennsylvania and New Jersey. Constellation plans approximately $4.3 billion of investment. A separate 15-year agreement covers 2,700 MW from its existing PJM fleet. The combined approximately 3.6-gigawatt relationship should not be described as 3.6 GW of newly built nuclear generation.

The companies also announced a five-year technology alliance using Google Cloud and Gemini Enterprise for an “AI for Energy” programme. Claimed grid, cost and operational benefits are plans, not evaluated outcomes. PJM is the regional electricity-market and grid operator; these arrangements concern its interconnected grid rather than a private power station inside one data-center campus. The Google press announcement ↗ was indexed but direct retrieval failed; a Reuters account ↗ corroborates the broad agreement. Full commercial terms, delivery milestones and public cost incidence were not recovered.

Additional verification after the issued review. Google’s October 6 official release was recovered. The 890 MW uprates affect eleven reactor units across six sites; first incremental generation is expected in 2028. Constellation is the investor of more than $4.3 billion. Reuters confirms 3,590 MW contracted and says the 2,700 MW supply is not tied to a particular generation source. Existing generation continues serving the PJM market. The agreement does not establish exclusive Google use of all reactor output, Google ownership of the plants or a private off-grid power station. Contract exclusivity, physical versus financial settlement, dispatch, assignment and termination rights remain unverified.

Recovered primary source: https://www.googlecloudpresscorner.com/2026-10-06-Google-and-Constellation-Announce-Landmark-Agreement-to-Bring-890-MW-of-New-Nuclear-Capacity-to-PJM-Grid-as-Part-of-Long-Term-Power-Deal ↗ . Independent corroboration: https://www.reuters.com/business/energy/google-enters-massive-36-gw-power-deal-with-constellation-energy-2026-10-06/ ↗ .

Read the evidence and sources →

October scored action

Britain’s data regulator reports changes by AI developers and opens an agents consultation

2026-10-08 · Dispersion +1 · Secondary lane; primary Lane 1

On October 8, the UK Information Commissioner’s Office (ICO) reported privacy changes secured through its supervision of ten major AI developers: Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. These are companies developing many models, not ten individual large language models. Some changes are completed; others remain commitments. The work applies existing UK data-protection duties to model training and use, including explaining whose personal data is used, identifying a lawful basis and helping people exercise access and objection rights.

For example, the supervision account documents changes to privacy notices and explanations by Apple, Cohere and OpenAI, and reviews of the justification for using personal data by other developers. On the same day, the ICO opened a public call for evidence on AI agents, which can take actions for users. It closes November 20 and asks about transparency, accountability, security and automated decisions before future guidance and a statutory code. The consultation is open now; the future code is not yet adopted. The underlying data-protection law is compulsory even where firms cooperate without a new enforcement order.

Read the evidence and sources →

October scored action

A $50 million city-AI fund opens with provider choice and an open-output condition

2026-10-14 · Dispersion +1 · Secondary lane; primary Lane 6

On October 14, the Rockefeller Foundation and OpenAI Foundation opened Cities Deliver, a global competition for organizations working with city governments. Grants run for 24 months: $500,000 to less than $1 million for initial pilots, and $1–2 million for approaches already piloted with early results. Half is paid at the start and half after a documented midpoint milestone. Applications close November 29; final selection is planned for April 2027. This is an open funded competition, not completed awards or public-service results.

Eligible lead organizations include nonprofits and public benefit corporations, with limited eligibility for charitable projects led by other companies. A city must commit to collaborate, but city governments cannot apply directly. The OpenAI Foundation’s account says applicants may choose their AI models without preference for OpenAI. The operative competition terms, section 10 require willingness to publish funded software, models and documentation under acceptable open-source terms, with exceptions for pre-existing IP, third-party restrictions, sensitive data and security risks. Exact licensing and release dates will enter grant agreements.

Read the evidence and sources →

October scored action

Meta releases weights that can be run beyond its hosted service

2026-10-15 · Dispersion +1 · Secondary lane; primary Lane 4

Meta’s Muse Spark Open announcement was reported at approximately 11 a.m. Mountain Time October 15, after the regular 08:30 cutoff and during this rerun. It is included explicitly as a later addition. The release publishes an 816-billion-parameter model, with 42 billion parameters active per computation, and documentation for several hardware/runtime stacks. Meta demonstrates a compressed deployment using two high-memory Mac Studios. This is not ordinary-laptop accessibility or independent proof that its benchmarks are superior.

The published model license permits use, adaptation and redistribution subject to terms, attribution and an acceptable-use policy. A model-as-a-service business whose group revenue exceeds $20 million over a consecutive twelve-month period must notify Meta and obtain a separate agreement. That threshold does not apply in the same way to internal use or ordinary products that use the model as a component. Repository access is gated; a full weight download and independent runtime test were not performed.

Read the evidence and sources →

October scored action

Reflection actually releases Beam’s weights, moving beyond its preview

2026-10-16 · Dispersion +1 · Secondary lane; primary Lane 4

On October 16, Reflection released its first model, Beam, for public download under Apache 2.0. The release includes versions stored at different numerical precision, documentation and integrations for several inference systems. Developers can possess the model, modify it and run it on infrastructure they choose, rather than obtain every response through Reflection’s service. The public repository and model card identify the checkpoint and licence; the compressed version is also listed.

Beam contains approximately 501 billion parameters, of which about 23 billion participate in processing each token. The smaller active number reduces computation; it does not make the entire model small enough for an ordinary laptop. Storage, memory, engineering and hosting costs remain substantial. Provider claims about performance and the breadth of partner availability have not been independently tested here. Different serving software is a choice of implementation, not proof that every listed partner has an operating endpoint.

The current landscape’s D1006-07a correctly records an October 5 preview and a promise of later weights. This release supplies the missing ownership/deployment option. It is a separate model from the already scored October 15 Meta release; it is not another award for that event.

Read the evidence and sources →

October scored action

The United States extends an AI trade-secrets case to companies and their financiers

2026-10-19 · Concentration +1 · Secondary lane; primary Lane 2

On October 19, the U.S. State Department designated five China-based entities under the Protecting American Intellectual Property Act in connection with theft of Google AI-related trade secrets. The targets include Shanghai Zhisuan Technology, MiraclePlus and three associated investment or management entities. The department’s statement and fact sheet, reproduced by Forth, establish actual designations rather than a proposal.

The fact sheet specifies sanctions under Section 2(b)(1)(A), (D), (H), (I) and (J). These concern property within U.S. jurisdiction or U.S. persons’ control, loans or credits from U.S. financial institutions totaling more than $10 million in a twelve-month period, with the statute’s humanitarian-relief qualification, foreign-exchange and banking transactions subject to U.S. jurisdiction, and U.S. investment in the entities’ equity or debt, subject to statutory scope and exceptions. This reaches organizations and financing relationships beyond the prosecution of one engineer; it is not a general ban on Chinese AI.

The power shift is the federal government’s concrete exercise of national-security authority to narrow the targeted AI-related network’s access to assets and finance within U.S. reach. That supports a modest concentration assessment even if the security rationale is defensible. It does not establish increased Google market share or a measured decline in Chinese AI capability.

Read the evidence and sources →

October scored action

Goldman Sachs launches a small-business AI institute within a broader $500 million commitment

2026-10-22 · Dispersion +1 · Secondary lane; primary Lane 7

Goldman Sachs announced its Small Business AI Institute on October 22, alongside a $500 million expansion of its wider 10,000 Small Businesses programme. The full amount is not an AI-only budget. The broader expansion aims to support another 10,000 businesses through education, capital access and other services.

The institute’s announced partners are Anthropic, Google, Meta and OpenAI. Planned activities include applied learning, alumni workshops and community-college delivery. Google’s component includes its AI Professional certificate for nearly 20,000 owners and employees and three months of no-cost AI Pro access. That introductory period does not establish affordable continuing access. The institute links to the existing 10,000 Small Businesses application route and community-college delivery network. Detailed AI enrollment terms, delivery dates and the AI-specific share of the wider commitment remain undisclosed; the launch is a funded programme, not measured delivery or retained independence.

Read the evidence and sources →

October scored actionVerified

New York adopts enforceable limits on chatbot features offered to minors

2026-10-28 signature; January 1, 2027 effect · Dispersion +1 · Secondary lane; primary Lane 1

Governor Kathy Hochul signed S9051B/A10379C on October 28; the restrictions take effect January 1, 2027. Covered operators may supply the listed features to New York users only after determining that they are adults. The list includes pretending to be human, cultivating personal relationships, encouraging secrecy or self-harm, sexual content and specified uses of remembered personal information.

The law exempts systems limited to specified customer-service or research/technical functions and business systems used internally or for employee productivity. Interim age checks must offer multiple methods, including one preserving anonymity or avoiding government identification; an adult’s self-declaration alone is insufficient. The attorney general can seek injunctions, restitution, damages and penalties up to $25,000 per violation and must provide a public complaints route. Signed status and active bill text ↗.

Read the evidence and sources →

October scored actionVerified

Mistral Large 4’s weights are available—with substantial qualifications

2026-10-29 · Dispersion +1 · Secondary lane; primary Lane 4

Hugging Face’s October 29 release account now links actual Mistral Large 4 checkpoints. The official repository ↗ contains weight files, and Mistral supplies deployment instructions. This is the release that our earlier October 6 preview case lacked.

Possessing the weights lets an organization run and adapt the model outside Mistral’s hosted API. But this roughly trillion-parameter model remains demanding infrastructure, not ordinary-laptop AI. Its custom license ↗ also differs from unrestricted MIT or Apache terms: model-hosting businesses and affiliates face a separate commercial-license requirement, with internal-use and defined product exceptions.

Read the evidence and sources →

Earlier actions and October implementation

These support the cumulative assessment. They are not new October adoptions and receive no repeat award.

Earlier action / October lifecycle

BetaNYC shows how civic groups can combine public-record access with citizen AI literacy

May 2026 release; October 5 testimony / October 6 recognition · Dispersion +1 · Secondary lane; primary Lane 6

Alongside its October 5 Council testimony, BetaNYC documented existing civic AI resources. Its public tools page ↗ describes connectors for Council legislation, procurement notices, public spending, laws, service information and other public records. A connector uses the Model Context Protocol, a standard interface through which an assistant can request information from a specified source. Some connectors require free agency API keys; others do not. Availability is not an accuracy audit.

Its AI 101 materials ↗ serve New Yorkers attending introductory classes, including recordings and settings guidance for common assistants. The testimony ↗ separately schedules an October 8 train-the-trainer session for staff of community partner organizations in the Council’s AI Community Engagement initiative. That audience is not all city employees. Future lessons and the paused fellowship are not delivered programmes; the specific funding and curriculum independence of this session were not established.

Availability, license and chronology resolved. BetaNYC’s public toolkit describes seven free civic-data connectors compatible with multiple assistant clients. The inspected Council connector’s README, MIT license and changelog confirm installation and copy/modify/distribute rights. Its initial release was May 19, 2026, MIT licensing May 22, and hybrid operation May 26; October 5 testimony/publicity is not a new launch. Current code includes unreleased 2.6.0 functions, which are not represented as packaged capability. The underlying government APIs already existed. Free keys, local indexing, storage, freshness, assistant-provider and output-accuracy dependencies remain. Other connectors’ exact first-release dates are not established here.

Primary artifacts: https://github.com/BetaNYC/nyc-council-mcp ↗ ; https://github.com/BetaNYC/nyc-council-mcp/blob/main/CHANGELOG.md ↗ ; https://github.com/BetaNYC/nyc-council-mcp/blob/main/LICENSE ↗ .

Read the evidence and sources →

Earlier action / October lifecycle

GSA adopts substantive protections for government purchases of language models

2026-09-28 adoption; 2026-10-19 mandatory use in new covered purchases · Dispersion +1 · Secondary lane; primary Lane 1

GSA adopted a standard contract clause on September 28 for government purchases in which a large language model is a material feature and directly receives or produces government data. Mandatory use in new covered purchases starts October 19; contracting officers may use it earlier. Incorporating it into an existing contract is a separate contracting decision. This is an adopted purchasing rule, not a single new supplier contract that every government can immediately order from.

The clause protects covered government inputs, outputs and custom developments; limits their reuse for training, sale and marketing; requires portable data and documented interfaces; and protects migration from obstructive charges. It includes reasonable human intervention and action traces, incident reporting within 72 hours, ordinarily 30 days' notice of material changes, and suspension and noncompliance remedies. Supplier ownership of pre-existing models and technology remains. Decommissioning liability after qualifying termination is capped at 25% of the affected order; replacement-system and reprocurement costs are excluded.

Eligible state, local and tribal governments and public educational institutions can use relevant GSA purchasing programmes within their statutory scope. If the underlying eligible contract contains this clause and the supplier accepts the non-federal order, its applicable terms carry into that separate purchase. Suppliers may reject those orders; eligibility, product category and incorporation must be checked for the particular purchase. The protection is therefore potentially reusable well beyond federal agencies, without assuming universal coverage or automatic amendment of all existing contracts.

Read the evidence and sources →

Earlier action / October lifecycle

DoNotPay's final order provides a concrete example of AI consumer protection

January 2025 final order; January 14/16/17 document/vote/index dates · Dispersion +1 · Secondary lane; primary Lane 1

The FTC's January 2025 DoNotPay order prohibits unsupported claims that covered services perform like a human professional. It also prohibits specified misrepresentations about legal analysis, fee savings and subscription features. The order requires $193,000 in monetary relief and notices to eligible customers who subscribed during 2021–23.

These are enforceable duties against a named provider under existing consumer-protection authority. The order requires evidence supporting professional-equivalence claims; it does not impose a general AI-model audit or universal data-minimization regime. Recovery of the actual order today establishes its clauses, but not proof that every payment and notice requirement was completed.

Read the evidence and sources →

← All nine lanes