Loading...
Loading...
California issued Executive Order N-12-23 on Generative AI in September 2023, directing every state agency to conduct AI use case inventories, establish governance policies, and report to the Governor's office on AI deployments. That order was followed by specific CDT (California Department of Technology) guidance in early 2024 that established procurement requirements, risk classification tiers for government AI systems, and a mandate for algorithmic impact assessments on high-risk applications. The result is the most thoroughly documented state AI governance framework in the country — and also, paradoxically, one of the slowest environments for actually deploying new AI systems, because every step from use case identification through production deployment now requires documented compliance with the CDT AI Readiness Framework and approval under the CA-RAMP (California Risk and Authorization Management Program) cloud security process. Cal HHS — the California Health and Human Services Agency, which oversees Medi-Cal, CDSS, and CDPH — is the largest single operational AI deployment surface in state government, administering health and social services for 14 million Californians. The California Department of General Services manages the statewide technology procurement contracts through which most AI vendors enter the California government market. LocalAISource works with California government clients who need AI partners that have already navigated CDT's documentation requirements, hold CA-RAMP authorization or have a clear path to it, and can produce the algorithmic impact assessment artifacts that procurement now requires.
Updated June 2026
California's AI procurement environment is the most process-intensive in the country, and vendors who discover this midway through a contract negotiation consistently underestimate both the timeline and the documentation burden. CDT's AI governance framework, published in its final form in late 2024, classifies government AI systems into four risk tiers: Low (informational tools with no decision-making function), Moderate (decision-support tools reviewed by humans), High (tools that directly affect benefit eligibility, law enforcement, or employment), and Critical (systems in life-safety or constitutional-rights contexts). High and Critical tier systems require algorithmic impact assessments, independent technical audits, and ongoing monitoring reports — requirements that add 3–6 months to deployment timelines and $150,000–$400,000 in compliance work to system budgets. CA-RAMP, the state's cloud security authorization program modeled on FedRAMP, adds a parallel authorization track for cloud-hosted AI services. Vendors who hold FedRAMP Moderate authorization receive expedited CA-RAMP review, but full CA-RAMP authorization for a new platform still takes 9–18 months. DGS's statewide technology contracts — the Software Licensing Program (SLP) and Cloud Technology Solutions (CTS) contract vehicles — provide pre-authorized access to a catalog of AI tools from major vendors (Microsoft, Google, Salesforce), but specialized AI vendors outside that catalog must run standalone procurements through either DGS or the California Multiple Award Schedules (CMAS) vehicle. The practical implication for AI vendors is that California government is not a market for fast sales cycles. The agencies with the most active AI programs — CDT itself, Franchise Tax Board, Employment Development Department — have internal AI teams that move faster on proof-of-concept work but require full procurement compliance before production deployment. Ask any California government AI program manager and they'll tell you: the algorithm documentation requirements have killed more pilots than the technical challenges have.
The California Health and Human Services Agency oversees a portfolio that includes Medi-Cal (12.5 million enrollees, $120B+ annual expenditure), CalFresh (4.5 million recipients), CalWORKs, and CDPH public health programs — making it the largest state social services enterprise in the country. Cal HHS's AI Strategy, released in 2024 under the umbrella of the Governor's GenAI EO, identifies five priority use case clusters: eligibility and enrollment automation, case management support, fraud and program integrity, public health surveillance, and workforce support tools for caseworkers. Medi-Cal's AI deployments are the most mature. The California Department of Health Care Services (DHCS), which manages Medi-Cal, has operated an ML-based provider fraud detection system since 2019 through a contract with Cotiviti — a system that cross-references provider billing patterns against the Medicaid Fraud and Abuse Unit's investigation database and generates risk scores for audit prioritization. DHCS reports that the system identifies 30% more high-risk providers per audit staff hour than the prior statistical sampling approach. The 2024 upgrade added behavioral health billing analysis, which had been excluded from the original model due to data sensitivity concerns that were resolved through a formal data use agreement with the California Office of Health Information Integrity. CDSS, which administers CalFresh and CalWORKs, deployed NLP-assisted document processing for benefit renewal packets in 2023 — a direct response to the post-COVID Medicaid unwinding backlog that California shared with other states but experienced at orders of magnitude larger scale. The system processes income verification documents in 15 of California's 58 counties, with statewide expansion planned for FY2025. The remaining 43 counties are on a staggered rollout schedule determined by county welfare department IT infrastructure readiness, which varies enormously between Alameda County (highly mature, early adopter) and rural Trinity County (minimal IT infrastructure).
The California Franchise Tax Board administers state income tax for 40 million filers and has one of the most sophisticated government data analytics programs in the country. FTB's ML-based compliance gap analysis, which cross-references federal IRS data, real estate transaction records from CORELOGIC, and financial institution data under the state's backup withholding reporting requirements, has been refined over a decade. The system identifies taxpayers with likely unreported income at a rate that FTB's annual reports credit with $1.5B+ in additional tax collections since 2017. FTB is now evaluating NLP tools for automated tax document intake — processing the 60 million paper and electronic forms FTB receives annually — with a pilot planned for the 2025 tax year that would auto-populate return data for taxpayers who opt into a state-prepared return program. The California Employment Development Department manages unemployment insurance benefits for a workforce of 19 million. The COVID-19 pandemic exposed catastrophic weaknesses in EDD's fraud detection infrastructure — the state paid an estimated $30B+ in fraudulent UI claims in 2020–2021, a loss that prompted a state audit and a technology modernization mandate. EDD's current fraud detection platform, rebuilt with input from a Governor-appointed Strike Team, uses ML-based identity verification (including document authenticity scoring and cross-agency identity matching) and behavioral analytics to detect bot-driven application patterns. The rebuilt system flags roughly 8% of new UI claims for human review, compared to the 0.3% fraud detection rate under the pre-pandemic rules-based system. For policy analysis, the California Department of Finance's Forecasting and Data Analytics unit uses ML-enhanced revenue forecasting models that have reduced the annual budget estimate error rate by 12% since 2022 — a meaningful improvement in a state where budget forecasting errors translate directly to mid-year spending cuts or unexpected surpluses. The Little Hoover Commission, California's independent government oversight body, published an AI governance review in 2024 that recommended the state establish an independent algorithmic accountability office — a recommendation the Governor's office has not yet acted on but that CDT has incorporated into its FY2025 policy agenda.
Strategic planning for AI adoption, readiness assessment, and roadmap development
Workflow automation using AI, including Make.com-style automation and RPA
Predictive models, data analysis, and ML pipeline development
Text analysis, document automation, sentiment analysis, and language processing
EO N-12-23 requires state agencies to complete use case inventories for all current and planned AI deployments, adopt CDT's AI governance framework including risk tiering and algorithmic impact assessments for high-risk applications, and report annually to the Government Operations Agency. It does not prohibit generative AI use but requires that any system affecting consequential decisions (benefits, law enforcement, employment) undergo independent technical review. CDT's implementing guidance, published in 2024, added specific procurement requirements: vendors must provide model cards documenting training data, performance metrics by demographic group, and limitation disclosures. Agencies that deployed AI before the EO had to complete retroactive documentation by June 2024.
CA-RAMP authorization is required for any cloud service hosting sensitive state data — which includes most government AI systems that process PII, health data, or financial records. Vendors with FedRAMP Moderate authorization receive expedited CA-RAMP review (typically 4–6 months). Vendors without FedRAMP authorization face 12–18 month CA-RAMP timelines. The practical shortcut is using a CA-RAMP-authorized cloud platform (Azure Government, AWS GovCloud, Google Cloud for Government) as the hosting environment and deploying the AI application within that authorized boundary — this shifts the authorization burden to the platform rather than the application vendor. DGS's Cloud Technology Solutions contract includes pre-authorized access to AI services within those platforms.
Cal HHS has deployed ML-based provider fraud detection through DHCS (Medi-Cal), NLP document processing for CalFresh and CalWORKs renewals through CDSS, and syndromic surveillance analytics through CDPH. The federal funding basis is primarily CMS Enhanced Match (90/10 federal match for MMIS improvements that include AI), which covers most of DHCS's AI investment. CalFresh AI tools are eligible for USDA FNS's 50/50 administrative match, which Cal HHS uses. The algorithmic impact assessments required by CDT add compliance costs that are not federally reimbursable, which is a budget tension several Cal HHS agencies have flagged.
DGS manages the Software Licensing Program and Cloud Technology Solutions vehicles, which cover major AI platforms. Specialized AI vendors must either enter the California Multiple Award Schedules (CMAS) through an application process (typically 3–6 months) or respond to agency-specific RFPs. CMAS qualification is the standard path: it requires a valid GSA Schedule 70 (IT services) or equivalent, financial statements, and a California Secretary of State registration. Once on CMAS, vendors can compete for any state agency contract without additional competitive procurement for transactions under $249,999. Above that threshold, a competitive process is required regardless of CMAS status.
EDD's rebuilt fraud detection platform, operational since late 2022, uses ML-based identity verification that cross-references Social Security Administration death master files, commercial identity databases, and cross-agency address records to flag suspicious applications. Behavioral analytics detect bot-driven application patterns — volume, timing, and data-entry characteristics that differ from human applicants. EDD's OIG reported in 2023 that the rebuilt system identified 12x more fraudulent claims per review staff hour than the pre-pandemic system, and that fraud as a percentage of total UI claims paid dropped from an estimated 10%+ in 2020–2021 to approximately 1.2% in 2023. The improvement is real but the comparison is partly a function of how catastrophically the pre-pandemic system had failed — the 2020 baseline was essentially no fraud detection.
Get discovered by California businesses looking for AI expertise.
Get Listed