California Governor Gavin Newsom and artificial intelligence company Anthropic announced a new agreement. California government agencies at all levels will use Claude, an AI chatbot developed by the company, at a discount and receive supporting training and technical support. The move comes as the business community struggles with high enterprise subscription fees for large-model tools and state governments look to lower-cost AI services to improve efficiency.
According to a press release issued by the governor's office, this agreement covers all state-level agencies and local governments. Relevant departments can deploy Claude in scenarios such as document writing and information analysis to assist civil servants in completing their daily work. Anthropic will provide system usage training and technical support to government employees to help them embed the AI tools into their existing workflows.
Newsom emphasized in the statement that the role of artificial intelligence is to "assist rather than replace" government work. AI should help staff advance tasks faster, solve problems more effectively, and provide better services to the people of California. He proposed that the application of AI should improve efficiency while ensuring that the quality and accountability of public services are not undermined.
The partnership continues an executive order signed by Newsom in March to accelerate the adoption of artificial intelligence within government to "make government more efficient" while establishing stricter safety standards. At the time of the order, Newsom had said that while some in Washington were developing policies and entering into contracts "under the shadow of abuse," California was moving forward with the use of AI "the right way."
In contrast to the California state government's continued closer relationship with Anthropic, the U.S. federal government's relationship with the OpenAI competitor continues to be tense. Earlier this year, Anthropic clashed with the U.S. Department of Defense over a contract that would have allowed the department to deploy Claude for any "lawful purpose." Anthropic wanted the contract to explicitly exclude uses such as using its technology to spy on Americans or deploy autonomous weapons without human oversight, but Defense Secretary Pete Hegseth refused. Eventually, the Department of Defense signed a cooperation agreement with OpenAI and further classified Anthropic as a "supply chain risk," prohibiting the company from working with other Pentagon contractors.
Although Anthropic was labeled a "supply chain risk" at the federal level, California did not take this into account when negotiating state contracts. Chris Given, California's chief information officer and director of technology, told POLITICO that the identification of "supply chain risk" was "not mentioned at all" when negotiating the contract with Anthropic.
As the cost of artificial intelligence infrastructure and computing power continues to rise, technology companies from OpenAI to SpaceX have developed their own chips to reduce their dependence on Nvidia. The cost issue of government procurement of AI services has become increasingly prominent. California’s discount agreement with Anthropic is seen as an exploration by local governments to actively intervene in the game between AI applications and costs. It not only responds to the business community’s general anxiety about AI subscription fees, but also attempts to provide a more controllable path for the digital transformation of the public sector.
From a political and technical perspective, California’s choice is in sharp contrast to the federal government’s attitude toward Anthropic, and it also reflects the differences within the United States on how to balance artificial intelligence safety, contract terms, and public interests. In the context of the Trump administration tightening some AI regulations and rolling back security protections, the Newsom administration on the one hand introduced Claude in a high-profile manner, and on the other hand emphasized the strengthening of AI usage rules and security boundaries through state-level executive orders, demonstrating a governance approach of "embracing technology but setting red lines."