White House finalizing classified Anthropic contract for NSA and U.S. spy agencies; $9B Nvidia Blackwell chip request cleared
The White House is finalizing a classified deal letting NSA and other U.S. spy agencies use Anthropic AI models, per New York Times. A carve-out blocks use against Americans' data. Separately, the administration cleared a $9B Nvidia $NVDA Blackwell chip buy for the agencies.
The White House is finalizing a classified contract that would let the National Security Agency and other U.S. intelligence agencies use Anthropic's advanced AI models for classified work, according to the New York Times. The deal arrives despite the Defense Department earlier labeling Anthropic a supply chain risk after the company and the Pentagon clashed over contractual language that could permit mass surveillance of Americans or autonomous weapons use. The pending contract includes a carve-out ensuring the model is not run on Americans' data, and the White House reportedly wants it to serve as a template for other AI vendors. U.S. officials say Anthropic's new Mythos model can run on previous-generation chips but performs more efficiently on the newest hardware. To address that, the administration has approved a $9 billion request for Nvidia $NVDA Blackwell chips for the spy agencies, and is reallocating $800 million to speed near-term computing purchases ahead of congressional approval. Anthropic is also continuing a court fight against the Pentagon's supply-chain-risk designation. Sources: New York Times, Arkansas Democrat-Gazette.