Anthropic, Code for America pilot AI tools for SNAP eligibility support - Government Executive

Anthropic, the company that assures us its models are practically dripping with conscientious restraint, has teamed up with Code for America to pilot AI tools for SNAP eligibility. Because clearly the biggest problem with food assistance is that caseworkers have too much time to think. The partnership will deploy a 'SNAP Policy Navigator' – a chatbot designed to interpret labyrinthine federal regulations – in a handful of state offices. Because nothing says 'nutrition security' like letting a language model decide who eats.

The Navigator's job is to help caseworkers answer policy questions and review eligibility documents. In other words, it's a glorified search engine with delusions of expertise. Anthropic's Claude will parse the intricacies of income disregards and work requirements with the same unerring confidence it displays when inventing citations. When a harried caseworker queries about categorical eligibility for a family with mixed immigration status, the model will produce a response that is either correct, wrong, or – most likely – both at the same time. And since the model can't be sued, fined, or subjected to congressional hearings, the accountability will trickle down to the human who trusted it.

This is a pilot, so the damage will be contained – for now. But it's a classic foot-in-the-door strategy: start with low-stakes policy questions, move on to document review, then gradually automate eligibility determination itself. Code for America has a track record of well-intentioned projects that struggle to scale; Anthropic has a track record of models that are as reliable as a coin flip. Together, they are offering the government a deal it can't refuse: free AI trial, with the cost of failure borne by the most vulnerable. What could possibly go wrong? The answer, of course, is everything – but that would require an audit, and no one wants to spoil the pilot.