California Built Its Own AI Tool Nine Months Before the Anthropic Deal
California Built Its Own Government AI Tool Nine Months Before Signing the Anthropic Deal
On June 29, Governor Newsom announced a first-of-its-kind partnership making Claude the first AI productivity tool available statewide across all California agencies and counties. The headline was the 50% discount. The more interesting detail — buried in the press release — is that 2,800 state employees across 67 departments had already been using a state-built AI platform before the formal deal was signed.
How this was reported: the facts in this article come from the official Governor’s press release (gov.ca.gov, June 29 2026), read in full, cross-referenced against TechCrunch, The Next Web, and Awesome Agents for factual confirmation. Quotes attributed to named individuals are reproduced verbatim from the official press release or from named outlets as cited. No interviews were conducted. This article covers government AI policy — not product assessment. Wangdoo has covered related AI governance developments in the UAE and the EU’s CADA regulation.
What the Announcement Actually Says
The California-Anthropic agreement, announced June 29 2026, makes Claude the first AI productivity tool available to all California state agencies through the state’s new Statewide Information Technology Shared Services (SITeS) portal — a centralised procurement channel managed by the California Department of Technology. The deal gives every state agency access at a 50% discount from standard pricing, with the same discount extended to cities and counties across the state. Free workforce training and direct technical assistance from Anthropic engineers are included. No total contract value was disclosed in the announcement or the press release.
ABC7 coverage of California Governor Newsom’s March 2026 executive order on AI accountability — the policy framework that preceded the Anthropic SITeS deal. Not a Wangdoo production.
That last point is worth noting. Secondary coverage described this as the largest US state government AI deployment in history — the Governor’s own press release does not use that phrase, describing it instead as “a first-of-its-kind collaboration.” Whatever the precise scale, the financial terms are not public.
The Poppy Detail Nobody Foregrounded
The press release describes several existing Claude deployments within California government before the formal deal was signed. The DMV is using it for customer service and wait time reduction. The Department of Healthcare Services — the largest Medicaid agency in the country — is using it for internal caseworker workflows. The California Department of Technology and the Office of Emergency Services are using Claude Security and Claude Code for scanning, triaging, and patching state code.
All of those were mentioned in secondary coverage. What received far less attention is Poppy.
Poppy is a state-built AI assistant — named after California’s official state flower — designed specifically for government workflows. It was not procured from Anthropic. It was built by California state workers themselves, running on a vendor-agnostic platform that provides access to multiple AI models — Claude among them — through a single secure interface, with pre-built query templates tailored to common state business needs. CDT’s own Poppy page confirms the pilot began September 29 2025 with more than 2,800 state employees across 67 departments. Poppy launched statewide on July 1 2026 — after the formal Anthropic deal was signed.
This changes the character of the story. Most coverage framed the announcement as: California will now give its workers access to an AI tool. The more accurate framing is: California’s workers were already using AI, had already built something on top of it, and the June 29 announcement formalises and centralises what was already spreading organically from within the workforce.
What Poppy actually is
Poppy is not a rebranded Claude interface. It is a vendor-agnostic government workflow platform built by state workers, providing access to multiple AI models — including Claude — through a single secure, state-managed interface. Pre-configured query templates are built around the specific tasks California government employees perform every day — drafting regulatory communications, summarising case files, responding to constituent inquiries, and similar work. The design philosophy — built by the workers who use it, for the tasks they actually do — is different from a top-down enterprise AI procurement, and it is why the pilot reached 67 departments before any formal vendor agreement existed.
The Federal Government Context
The political backdrop to this announcement is explicit and worth understanding factually rather than through either side’s framing.
Earlier in 2026, Anthropic and the US Department of Defense entered contract negotiations. The DoD sought the ability to deploy Claude for any lawful use. Anthropic sought to carve out explicit protections preventing deployment for mass domestic surveillance and autonomous weapons systems without human oversight. Defense Secretary Pete Hegseth refused those terms. The Pentagon signed a deal with OpenAI instead and subsequently declared Anthropic a “supply-chain risk,” a designation that prevented Anthropic from working with other Pentagon contractors.
On March 27 2026, US District Judge Rita Lin in San Francisco issued a preliminary injunction blocking enforcement of the supply chain risk designation, writing that the measures “appear designed to punish Anthropic” and constituted “classic illegal First Amendment retaliation.” The case is not resolved — on April 8, a separate federal appeals court in Washington DC denied Anthropic’s request to block the designation at the appellate level. The legal status remains in litigation. The preliminary injunction is confirmed from CNN, The Hill, and CNBC court reporting.
California’s CIO and Department of Technology director Chris Given told Politico that the federal supply-chain risk designation “just didn’t come up” during the state’s contract negotiations with Anthropic. The state proceeded on its own assessment of the vendor rather than deferring to the Pentagon’s designation.
Engaged California — The Part That Has No Precedent
One deployment in the announcement has received almost no specific coverage: Engaged California. This is described by the Governor’s office as “a first-in-the-nation deliberative democracy platform” that uses Claude to allow Californians to engage with policy questions directly. The platform is intended to give citizens a structured channel to contribute to policymaking — not just submit comments, but engage in something described as deliberative dialogue mediated by AI.
What that means technically — how the AI structures the deliberation, what outputs feed into policy processes, and what safeguards prevent the AI from steering outcomes — was not specified in the press release. The ambition is stated. The mechanism is not. For a state of 40 million people, a deliberative democracy platform powered by AI at scale is genuinely novel and genuinely uncharted. The June 29 announcement is not the place to evaluate it — but it is worth flagging as the deployment that warrants the most scrutiny as it develops.
What the SITeS Portal Changes
The SITeS portal is not only about Claude. It is California’s centralised AI procurement infrastructure — a single channel through which state agencies can access vetted AI tools with transparent pricing. Claude is the first tool available through it, but the portal’s existence signals a shift in how California intends to manage AI procurement going forward: centrally, with visible pricing, rather than through fragmented agency-by-agency contracts.
For technology vendors, this matters. A centralised procurement portal that gates access to the California state government is a significant structural development. Any AI company seeking public sector contracts in California will need to navigate SITeS rather than approaching individual agencies. The March 2026 executive order that established the procurement standards — requiring vendors to demonstrate responsible practices on bias, civil rights, and misuse prevention — defines the entry criteria for that portal. The Anthropic deal is the first to emerge from those standards. It sets the template.
What Remains Unknown
Several consequential facts are absent from the public record.
The total contract value has not been disclosed. The specific Claude product tiers available through SITeS have not been specified — whether agencies access a standard commercial version or a specific government deployment is not in the public announcement. The data handling terms — how California government data processed through Claude is stored, retained, or used — are not described in the Governor’s press release. These are the questions that matter for a public institution handling Medicaid case files, cybersecurity incident data, and citizen policy submissions.
Anthropic’s own AI usage policies govern commercial deployments. Whether the California deal includes specific contractual modifications to those policies — data retention terms, audit rights, output logging — is unknown from the public documents.
My Take — Mr Wangdoo
Reading the primary source carefully rather than the secondary coverage: the Poppy detail is the editorial lead this story deserved and did not get. The bottom-up adoption pattern — 2,800 workers across 67 departments building and using something before any formal procurement existed — is the signal that AI adoption in this government context is not being driven primarily by executive mandate. It is spreading because workers find it useful enough to build tools on top of it themselves. That is a different and more robust form of adoption than a procurement announcement followed by a training rollout that employees treat as optional.
The missing contract value is the thing I keep returning to. The largest US state government AI deal in history, and the financial terms are not public. For a public institution spending public money, that gap between the scale of the announcement and the transparency of the terms is worth tracking. It may be disclosed later through California’s normal public records processes. It may not be. Either way, the absence is notable.
The Engaged California deliberative democracy platform is the most genuinely novel part of this announcement and the least examined. Using AI to mediate structured policy dialogue at scale with 40 million residents, in a way that feeds into actual government decisions, has no real precedent. How it works and what guardrails exist will matter considerably more than the discount rate on enterprise AI licences.
Frequently Asked Questions
What is the SITeS portal and what does it do?
SITeS stands for Statewide Information Technology Shared Services. It is California’s centralised AI procurement portal, managed by the California Department of Technology. It gives all state agencies — and, under the Anthropic deal, cities and counties — access to vetted AI tools at negotiated pricing. Claude is the first tool available through it. The portal is intended to standardise AI procurement across California government rather than allowing each agency to negotiate separately.
What is Poppy and how is it different from just using Claude?
Poppy is an AI tool built by California state workers, running on a vendor-agnostic platform with access to multiple AI models including Claude, built on state-managed infrastructure. CDT confirms the pilot began September 29 2025, with 2,800+ employees across 67 departments. It is purpose-built government workflow software, not a rebranded Claude interface. The June 29 announcement formalises the broader AI infrastructure that Poppy was already using.
Why did the Pentagon declare Anthropic a supply-chain risk?
The Pentagon and Anthropic negotiated a contract in 2026. The DoD sought permission to deploy Claude for any lawful government use. Anthropic sought to exclude deployment for mass domestic surveillance and fully autonomous weapons without human oversight. Defense Secretary Pete Hegseth rejected those terms, signed a deal with OpenAI instead, and subsequently designated Anthropic a supply-chain risk — preventing other Pentagon contractors from working with Anthropic. On March 27, US District Judge Rita Lin issued a preliminary injunction blocking enforcement of the designation, writing it “appears designed to punish Anthropic” and constituted “classic illegal First Amendment retaliation.” On April 8, a DC federal appeals court denied a separate Anthropic request to block the designation at the appellate level. The legal status remains in active litigation as of publication.
Is the California-Anthropic deal available to the public?
The 50% discount and free training apply to California state agencies and to cities and counties that opt in. Individual California residents do not receive discounted access through this deal. The SITeS portal is a government procurement channel, not a consumer service.
What is Engaged California?
Engaged California is described by the Governor’s office as a “first-in-the-nation deliberative democracy platform” that uses Claude to facilitate structured policy dialogue between California citizens and state government. The technical details of how the AI structures deliberation, what outputs feed into policy decisions, and what safeguards exist were not disclosed in the public announcement. It has been used in the development of at least one policy process. As a model for AI-mediated civic engagement at state scale, it has no direct precedent in the United States.
Sources
- Governor Newsom announces first-of-its-kind partnership with Anthropic — Office of the Governor of California, June 29 2026 (primary source, read in full)
- Anthropic and Gov. Newsom forge deal allowing California government to use Claude at half price — TechCrunch, June 29 2026
- Judge blocks Pentagon’s effort to ‘punish’ Anthropic — CNN, March 26 2026 (District Judge Rita Lin preliminary injunction)
- Judge blocks Pentagon’s supply chain risk designation for Anthropic — The Hill, March 27 2026 (full ruling details)
- California gives all state agencies access to Claude at half price — The Next Web, June 29 2026
