The White House is putting the finishing touches on a deal with Anthropic that would let the National Security Agency (NSA) and other U.S. intelligence agencies tap into Claude AI for top-secret analytical work, according to a New York Times report. Simultaneously, the White House signed off on a $9 billion purchase of NVIDIA Grace Blackwell chips to build a dedicated AI compute infrastructure for the intelligence community.
From 'Supply Chain Risk' to Intelligence Core
Just months ago, the Pentagon had flagged Anthropic as a 'supply chain risk,' placing it alongside AI services with Chinese ties on military procurement lists. Now the same company — built on a 'responsible AI' and 'safety-first' ethos — is set to become a core tool for spy agencies. Anthropic co-founder Dario Amodei had repeatedly opposed military uses of AI. Yet facing national security pressure and a potential multi-billion-dollar government contract, the firm once seen as an ethical benchmark crossed its own red line.
How the Big Three AI Labs Differ on Military Ties
U.S. top AI labs take sharply different stances on military cooperation. OpenAI quietly altered its usage policy in early 2024, replacing 'no military and warfare use' with a vague 'comply with international law' phrase, soon followed by reports of Pentagon projects. Google DeepMind, despite internal employee pushback, keeps participating in U.S. military medical diagnosis and logistics optimization programs. Anthropic had been the most hawkish, even refusing to develop model features that could be used for surveillance. This deal effectively demolishes its 'ethical firewall.'
$9 Billion NVIDIA Order and Taiwan's Key Role
The structural push from hardware runs deeper. The $9 billion NVIDIA chip order approved by the White House will mostly go to new 'intelligence-dedicated AI data centers.' These chips, built on the Grace Blackwell architecture, are designed for massive-scale inference and training — clearly meant for running Claude-level giant models. Manufacturing and packaging these chips hinge on Taiwan's semiconductor supply chain. TSMC's advanced packaging and CoWoS capacity are the lifeline for NVIDIA's flagship chips; Taiwan is already a de facto technological pillar in the AI arms race, not a bystander.
Privacy and Security Red-Line Debate
The deal has ignited fierce debate inside the U.S. over security red lines. Privacy advocacy groups warn that giving the NSA access to cutting-edge AI will supercharge mass surveillance and data analysis, potentially infringing on civil rights. Some lawmakers worry that if Anthropic's AI model shows bias or gets infiltrated by adversaries, the entire intelligence system could face systemic risk. Ironically, the very unit that listed Anthropic as a risk months ago is now leading the charge to embrace it.