Should government be doing deals like this? Yes. The free-market complaint about government has never been that it uses modern tools. Rather, the complaint is that government so often refuses to use them. Californians wait in DMV lines, stay on hold for hours with state agencies, and get stuck in permitting purgatory because the state runs on systems the private sector abandoned a generation, or more, ago. If AI helps a caseworker clear a backlog or a clerk answer a question in minutes instead of weeks, taxpayers win. Efficiency is not a partisan value. It is what we are owed in these cases.
The question is not whether but how, and how is where Newsom’s approach deserves scrutiny.
Claude will be the first, and for now the only, AI productivity tool offered to every agency through the state’s new shared services portal. First should never quietly become only. When government anoints a single vendor, competition withers, switching costs mount, and renewal prices climb. Notably, the announcement disclosed no total cost and no projected savings. A discount off an undisclosed number is not transparency, that’s marketing.
Then there is the whiplash. For years Sacramento has piled mandates on the AI industry, from SB 53’s disclosure regime to dozens of bills treating the technology as a menace requiring permission slips. Now the same governor declares AI trustworthy enough to draft documents, analyze information, and touch Californians’ data inside state government. If Claude can be trusted with the state’s own workflows, then the entrepreneurs and employers building with these tools deserve the same presumption of liberty the governor extends to himself.
Contrast that with the federal approach. The Trump administration’s General Services Administration did not pick a champion. Rather, America’s leading AI companies were invited to compete for the government’s business. OpenAI offered ChatGPT to federal agencies for one dollar, Anthropic matched at one dollar, and Google undercut both at 47 cents, all in service of an AI Action Plan built on removing regulatory barriers rather than adding to them. And when Anthropic sought to attach conditions to defense use of its models, the Pentagon walked away and signed with a competitor.
Whatever one thinks of that dispute, Washington negotiated from strength and preserved choice. Sacramento negotiated a partnership with a hometown favorite. The federal deals were not flawless, and one-dollar introductory pricing invites hard questions in year two. But the model, competition plus deregulation, is the right one.
The differences run deeper than procurement. Newsom has openly framed his deal as a rebuke of Washington’s approach to AI contracting. The president’s AI Action Plan treats artificial intelligence as a race America wins by unleashing its builders. Newsom’s “California way” treats it as a force government must first tame, then harness for itself. One philosophy trusts markets to discipline vendors and prices. The other asks taxpayers to trust its government overseers.
Californians should not fight AI in government. They should fight for it to be done the free-market way.
Legislators need to engage as well. They should demand publication of the full contract and its costs, insist that competing tools be added to the portal and rebid regularly, require agencies to report measurable savings, and pair every dollar spent on government adoption with an equal measure of regulatory restraint for the citizens who pay the bills.
Newsom found the right tool. Governments should buy the best technology America produces, and Claude is a serious product from a serious company. But governments should buy it the way markets work, through competition, transparency, and accountability. The danger was never the algorithm, but rather the monopoly instinct of the people procuring it.
Bartlett Cleland is a senior fellow in tech and innovation at the Pacific Research Institute.