No Single Mind: Why Artificial Intelligence Will Not Stay Centralized, Why a State AI Monopoly Would Move America Toward Socialism, and Why Security, Not Ownership, Is the Real Priority
This paper makes two arguments and draws one practical conclusion. First, under current technological and market conditions, artificial intelligence cannot be held by one controller for long. Open-weight models trailed the closed frontier by about four months in 2026, the price of a fixed level of capability falls by an order of magnitude or more each year, weights leak, rival laboratories extract capabilities through millions of queries, and export-controlled chips reach their targets anyway. AI already reaches about a fifth of US businesses and about half of all jobs, much of it through agents their owners run themselves. Second, because competition has not produced a single controller and on present trends will not, centralization would have to be imposed by the state, creating a state AI monopoly through ownership of the leading firms or discretionary licensing of who may train, publish, and use frontier models. That rebuilds the central planner of the socialist calculation debate and moves the United States toward state direction of a general-purpose means of production. At their latest reported valuations, buying the two leading laboratories outright would cost about $2.75 trillion and a majority stake about half that, before control premiums, and no purchase reaches foreign laboratories, open models, or private agents. China shows that a state can still direct many firms at once. The strongest case for centralization, that competing laboratories race and cut safety, is answered by shared safety and security rules rather than ownership. Centralization could also leave fewer employers and fewer new firms to open AI jobs, and it would be a strategic liability: one target for every adversary, one point of failure, and one set of blind spots shared by everything that depends on it. The paper concludes that the pressing risk is security: the failures on record are failures of authentication, containment, and access, and the answer is defense across many systems, not ownership of one. The deposit package contains the manuscript PDF, its LaTeX source, a README, and SHA-256 checksums. The paper uses no new data; all evidence comes from the 93 published sources in its reference list. Version 5. Answers an outside review and three adversarial reviews (economist, AI governance, security). Narrows the test of Section 2 to state ownership or discretionary allocation, so that safety rules open to every qualified applicant fall outside it, citing Mises's account of socialism under nominal private ownership. Explains in Section 6 why AI, as a decision and content layer, differs from electricity or GPS run as public monopolies, and limits the Hayek argument to allocation by permission. Marks the automation argument as conditional, citing Higgs on the ratchet of state power, and recasts the cost table as a $1.4 to $2.75 trillion range of fiscal outlay before control premiums. Adds the weakest-link argument for concentration, an answer to the claim that only the state can secure AI (the 2017 Shadow Brokers leak), state-grade security for plural private laboratories, the security costs of open weights, and warning signs of a decisive lead. Corrects the Ding case, the OpenAI and AISI incidents, Belfield's proposal, and China's filings, and adds eight sources (93 in all). An independent check found ten items, all fixed before deposit. Version 4: 10.5281/zenodo.24355536. Version 3: 10.5281/zenodo.24355133. Version 2: 10.5281/zenodo.24354586. Version 1: 10.5281/zenodo.24354168. Declaration on the use of artificial intelligence. The thesis, framework, and interpretation in this paper are the author's. An AI assistant (Claude, Anthropic) was used substantially as a tool under the author's direction for literature search, source checking, drafting, and typesetting. The arithmetic in Section 6.5 was checked by executed code. The author reviewed the final text and takes full responsibility for its content. Competing interests. The author is the founder of Guarino Industries LLC, which develops NINE, an offline software simulation of agents mentioned in the paper. The research was self-funded.
Authors
- Brian Guarino (ORCID: https://orcid.org/0009-0008-6705-8705)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.5281/zenodo.24354167
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- preprint