📰 Key Takeaways

California Governor Gavin Newsom has struck a deal with Anthropic, allowing all state agencies and local governments in California to use Claude at a discounted rate. Under the deal, government units not only get access to the Claude chatbot but also receive training and tech support from Anthropic. Newsom says Claude will primarily help state employees draft documents and analyze information, emphasizing that AI should assist humans rather than replace them—the goal being to help civil servants solve problems faster and deliver better service to California residents. This agreement builds on an executive order signed by Newsom in March, which aimed to accelerate government adoption of AI to boost administrative efficiency while maintaining stricter safety standards.

The backdrop to this is especially noteworthy: while California is drawing closer to Anthropic, the federal government’s relationship with the company has deteriorated rapidly. Earlier this year, Anthropic clashed with the U.S. Department of Defense over a contract—the DoD wanted to use Claude for any lawful purpose, but Anthropic insisted on explicitly excluding surveillance of American citizens and deploying autonomous weapons without human oversight from the agreement. Defense Secretary Pete Hegseth refused to accept these terms and instead signed with OpenAI. The federal government went further, labeling Anthropic a “supply chain risk,” prohibiting the company from working with any other Pentagon contractors. In response, California’s Chief Information Officer Chris Given told media that this “supply chain risk” designation was “never mentioned” during negotiations with Anthropic, showing that California is deliberately charting an AI governance course that differs sharply from the federal government.

💬 JudyAI Lab Perspective

The California-Anthropic deal sends a clear signal: AI’s “boundaries of use” are escalating from a technical setting to a governance choice—and where that line is drawn is determining who your partners are and who labels you a risk.

Anthropic insisted on explicitly excluding citizen surveillance and unsupervised autonomous weapons deployment from its contract, only to be labeled a “supply chain risk” by the federal government—while simultaneously becoming California’s go-to partner. For us AI builders, this case proves one thing: “what this AI refuses to do” has become just as important as “what this AI can do.” One remark from California’s CIO is particularly worth noting—“the supply risk designation was never mentioned during negotiations”—which suggests local governments are deliberately cutting themselves off from federal procurement decisions. An AI vendor’s stance on boundaries is actually reshaping the entire government market. Clearly defining your product’s boundaries might drive away some big clients, but it’ll precisely attract partners who share your values.

Why not write a clear “won’t do” list for the AI tool you’re building? That list might earn more trust from the right clients than your feature descriptions ever could.

📅 Original Source

🔗 Further Reading