AskCA: California Puts Claude to Work as a Digital Assistant for 39 Million Residents

On September 9, 2026, Governor Gavin Newsom introduced AskCA, a new AI-powered digital assistant designed to help Californians navigate state and local government services through a single entry point. Built by the state’s Office of Data and Innovation (ODI) in collaboration with Anthropic — whose Claude model underpins the tool’s knowledge base — AskCA represents the most visible deliverable so far from the Governor’s March 2026 Executive Order on artificial intelligence. And in a departure from the usual announcements, the state is shipping something usable on day one: a signup portal for volunteer beta testers, plus a live production feature for state job seekers arriving September 30.

From Wildfire Ashes to a Statewide Assistant

The origin story matters here. AskCA grew out of California’s response to the devastating January 2025 Los Angeles wildfires, when residents struggling to recover discovered just how hard it is to find help across more than 200 state agencies and departments. “When a Californian needs help, they shouldn’t have to become an expert in government to find it,” Newsom said in the announcement.

The design philosophy that emerged flips the traditional government portal on its head. Instead of organizing information around how Sacramento is organized — departments, bureaus, program names — AskCA is organized around life events: finding a job, starting a business, accessing family services, and recovering from a disaster. Jeffery Marino, Director of the Office of Data and Innovation, described the shift plainly: “This digital assistant moves from traditional web search to a vision of a service-integrated tool, explaining services, streamlining logins, and unifying department services into a single experience.”

What AskCA Actually Does

AskCA is not a chatbot bolted onto ca.gov. The state emphasizes three structural choices that distinguish it from consumer AI assistants:

• Verified sources over generative improvisation. Subject matter experts from across state government continually inform the responses, and answers are grounded in verified state sources to provide “clear and actionable guidance rather than simple chat functionality.” Stanford’s RegLab, led by professor Daniel E. Ho, is partnering with the state to “rigorously learn, evaluate, and improve such a system through rapid technological change” — an explicit nod to the concern that a static evaluation would be obsolete within months.
• Privacy by default. The CalHR deployment, at least, does not store personal information and requires no account.
• Human-centered testing before launch. Before today’s debut, the prototype was researched and tested with more than 140 fire recovery leaders and Altadena wildfire survivors, a dozen job seekers at a recent CalHR job fair, and over 100 general user testers.

The Anthropic Connection

The knowledge base behind AskCA was built “in collaboration with Anthropic’s Claude model,” with security support from the California Department of Technology. That formalizes a relationship that has been building all year: in June and July 2026, California struck a deal giving state agencies access to Anthropic’s tools at half price, along with workforce training and technical assistance. Anthropic also provides strategic guidance on the AskCA project itself.

For Anthropic, a high-profile, safety-scrutinized government deployment is a meaningful reference win — evidence that Claude can run in a domain where a hallucinated answer about, say, disaster relief eligibility has real consequences. For California, the partnership gives a state office access to frontier-model capabilities it could not build alone. The state is also quick to note the broader coalition involved: U.S. Digital Response and Stanford’s RegLab help measure performance through the California Innovation Council and Emerging Technology Accelerator, and a partnership with 211 — the community services hotline — extends AskCA’s reach to residents who may not reach for a web tool first.

First Production Use: Matching Skills to State Jobs

The most concrete piece of the launch is CalHR’s. On September 30, CalCareers.ca.gov will debut an AI-driven job-skills-matching feature powered by the same engine behind AskCA. Job seekers upload a resume and receive matching state job classifications — an approach early testers called especially helpful for those new to public service, where job titles like “Staff Services Analyst” famously map poorly to private-sector experience.

“CalHR is proud to be the first department to put the AskCA prototype into action to improve services for Californians,” said CalHR Director Monica Erickson. The feature launches as a pilot with full public access, allowing refinement in production while preserving privacy.

The Bigger Picture: Government Efficiency as AI Policy

Newsom framed AskCA as part of a broader efficiency agenda. His administration has already established the Office of Data and Innovation, overhauled the DMV, built the “Poppy” AI tool tailored for government use, launched the Emerging Technology Accelerator with pro bono expertise from UC Berkeley, Stanford HAI, Mozilla, and Nava Labs, and issued executive orders in 2023 and 2026 directing agencies to responsibly adopt generative AI with privacy, security, and civil-liberties guardrails. The Breakthrough Project, another initiative, claims $12 million in savings through pooled IT contracts.

California is effectively running two AI tracks at once: a regulatory one — the state has been among the most aggressive in scrutinizing frontier models — and an adoption one, where it behaves like any large enterprise betting on AI to cut administrative friction. AskCA is the clearest sign yet that the second track is producing citizen-facing software. Nick Maduros, Secretary of the Government Operations Agency, put the ambition in one line: “By creating a single entry point for government services, we are removing steps in the process and improving efficiency for Californians.”

What to Watch

AskCA is a prototype seeking beta testers — Californians 18 and older can sign up at ask.ca.gov — so the hard questions are still ahead. How will the state measure accuracy across hundreds of programs? Will Stanford’s evaluation framework catch failures before residents do? And can a tool built around four life events scale to the long tail of state services without collapsing back into the search-box problem it was designed to replace?

The September 30 CalCareers launch will be the first real-world test. If resume-to-classification matching works, expect more departments to follow; if it stumbles, AskCA joins the long list of government IT promises that shipped late and underdelivered. Either way, with 39 million residents as the user base, California has just become the largest live experiment in AI-assisted government service delivery in the United States.