The White House and Anthropic have reached an agreement under which the NSA will be fully integrated into Claude AI.

A few months ago, the U.S. Department of Defense labeled Anthropic a "supply chain risk" and even listed it alongside other Chinese-backed AI services on its military procurement list. However, a recent report by The New York Times reveals that the White House is currently finalizing an agreement with Anthropic, allowing the National Security Agency (NSA) and other U.S. intelligence agencies to use Claude AI for top-secret analytical work. At the same time, the White House has also approved a $9 billion NVIDIA Grace Blackwell chip project to provide the intelligence community with dedicated AI computing infrastructure.

The Shift from "Supply Chain Risk" to Intelligence Core

This agreement symbolizes a major shift in the US government's trust in AI. Previously, intelligence agencies were constrained by data classification levels and regulatory red lines, making it virtually impossible to directly use commercial cloud-based AI models. Anthropic, since its inception, has consistently touted "responsible AI" and "safety first," with co-founder Dario Amodei repeatedly and publicly opposing the use of AI for military purposes. However, faced with pressure from national security concerns and a potential multi-billion dollar government contract, this AI company, once considered a "moral benchmark," ultimately chose to cross the line.

Comparing the military stances of the three major AI giants

It's worth noting that the attitudes of top US AI labs towards military cooperation vary greatly. OpenAI quietly revised its usage policy in early 2024, removing the phrase "prohibited for military and wartime use" and replacing it with the vague statement "in accordance with international law," subsequently announcing project collaborations with the Pentagon. Google DeepMind, despite strong internal backlash, continued to participate in the US military's medical diagnostics and logistics optimization programs. Anthropic, previously the most hawkish, even refusing to develop model features that could be used for surveillance, now effectively declares the complete collapse of its "ethical firewall" with this agreement.

Taiwan's semiconductor industry plays a key role in NVIDIA's $9 billion order.

A deeper structural impetus comes from the hardware level. The vast majority of the $9 billion NVIDIA chip order approved by the White House will go to the newly built "AI Data Center for Intelligence." These chips use the latest Grace Blackwell architecture, designed specifically for large-scale inference and training, clearly intended to execute massive models at the Claude level. The manufacturing and packaging of these chips rely heavily on the key role of Taiwan's semiconductor supply chain. TSMC's advanced packaging and CoWoS capacity are the lifeblood of NVIDIA's flagship chip mass production; Taiwan is no longer a bystander in the entire AI arms race, but a substantial technological cornerstone.

The debate over the red lines of privacy and security

This agreement has also sparked a heated debate within the United States regarding security red lines. Privacy advocacy groups point out that the NSA's acquisition of licensing rights to state-of-the-art AI will further enhance its mass surveillance and data analysis capabilities, potentially infringing on civil rights. Some members of Congress worry that if Anthropic's AI models malfunction or are infiltrated by adversaries, the entire intelligence system will be exposed to systemic risks. Ironically, entities that listed Anthropic as a risk just months ago are now leading the way in embracing the system.

For Taiwan, this news is not merely a distant technological and political development. On one hand, Taiwan's semiconductor industry indirectly supports the hardware foundation for this intelligence program; on the other hand, with the full integration of AI into the US intelligence community, its ability to predict and analyze the movements of the Chinese People's Liberation Army will be greatly enhanced, and an "AI asymmetric advantage" in regional security is taking shape.

Related reports

NSA's Unauthorized Use of Anthropic Mythos: The Pentagon's Two-Faced Approach of Simultaneously Blocking and Allowing Its Release

Anthropic Report: The 2028 AI Hegemony Race – If the US Fails to Defend Its Computing Power Advantage, It May Be Overtaken by China.

A deep dive into the Anthropic account banning storm: a security religion, an AI civil war, and Claude's predicament amidst US-China decoupling.

The U.S. Department of Defense signed an AI cooperation agreement with seven tech giants, including OpenAI, Microsoft, and Nvidia; Anthropic refused to compromise and was subsequently excluded by Trump.