US Agencies Reportedly Using Anthropic’s AI Despite Pentagon Blacklist U.S. national security officials are locked in a paradox: Anthropic is formally treated as too risky to trust, yet its AI models are reportedly too capable to abandon on current hardware.

2019–2025: From concern to blacklisting

Over several years, Pentagon officials grew uneasy with Anthropic’s refusal to accept an “all lawful use” standard for its AI tools, worrying the company might cut off access over disagreements about how the military used its systems. Unlike some rivals, Anthropic insisted on red lines around mass domestic surveillance and weapons that fire without human involvement, arguing it couldn’t control models once deployed into classified environments.

Those tensions culminated in a Pentagon designation of Anthropic as a national security “supply chain risk,” effectively blacklisting the company as a security threat even as the government explored using its most powerful model, Mythos, for cyber defense.

May 2026: Courtroom clash over the blacklist

In mid‑May 2026, the Trump administration defended the blacklist in federal appeals court, arguing it was “unworkable” to rely on a vendor that might withdraw service based on “ideological” AI safety views. Judges pushed back, with one calling the move “a spectacular overreach” and saying she saw no evidence of malicious intent by Anthropic. Another judge highlighted the challenge posed by fast‑evolving, opaque AI models, noting that “AI three months from now will be totally different from the AI of today.”

Late May 2026: Chip shortage forces continued use

Days later, reporting revealed that despite the blacklist, the National Security Agency continues to use an advanced Anthropic model because “there is no alternative,” given a critical shortage of high‑end chips needed to run other frontier systems on top‑secret networks. The White House quietly approved a $9 billion emergency funding request to build specialized data centers for Nvidia’s Grace Blackwell superchips, aiming to end reliance on Anthropic and head off fears that China could gain a computational edge in intelligence operations.

The result is a stark contradiction: a U.S. company labeled a security threat has become, at least temporarily, indispensable to the very spy agencies that must mitigate that risk.

Continue reading https://foxvector.com