Code for America partners with Anthropic to build AI tools for SNAP caseworkers

Code for America partners with Anthropic to build AI tools for SNAP caseworkers

Code for America announced Thursday a partnership with Anthropic to integrate Claude, the AI assistant, into tools designed to help government caseworkers manage federal benefits programs. The nonprofit civic tech organization will start with the Supplemental Nutrition Assistance Program, creating a tool called the SNAP Policy Navigator.

The navigator gives caseworkers real-time access to federal, state and county SNAP guidance. It aims to reduce the administrative work tied to determining eligibility and interpreting policy rules.

Beyond the initial pilot, Code for America and Anthropic plan to expand Claude's role to include reviewing eligibility documents, answering policy questions and drafting plain-language communications for benefit recipients.

Why this matters for government operations

States face new financial pressure to improve accuracy in benefits administration. Under H.R.1, the budget reconciliation law passed last year, states now share the cost of administering SNAP based on their payment error rates and how accurately they determine eligibility and benefit amounts. Error rates above 6% trigger corrective actions and potential financial penalties.

State IT systems that handle benefits cases-known as integrated eligibility and enrollment, or IEE, systems-translate federal, state and local policies into computer code. A February report from the Digital Benefits Network at Georgetown University found that these systems often suffer from technological complexity, aging infrastructure and coordination problems across agencies.

The report identified AI as a tool that can accelerate policy implementation. Under what's called a "rules as code" approach, AI systems can interpret and apply rules directly, increasing transparency and reducing ambiguity in how policies get coded into software.

How Claude will be used

The SNAP Policy Navigator builds on Anthropic's Model Context Protocol, an open standard designed to ground AI responses in verified policy information rather than general knowledge. This distinction matters: caseworkers need answers tied to actual regulations, not plausible-sounding outputs.

States are already moving beyond AI experimentation toward broader deployment. Michigan's Department of Health and Human Services deployed an AI tool last March to help employees review cases more accurately. Maryland secured grants for AI projects to connect residents to SNAP and Medicaid services.

For government workers managing complex eligibility rules under tight timelines, the tool addresses a real operational constraint: caseworkers currently interpret dense policy manuals while processing cases. Faster access to accurate policy guidance could reduce processing delays and errors.

Learn more about Claude AI Courses or explore AI for Government to understand how these tools work in public sector operations.