Bloomberg TechnologyMay 22, 2026, 01:38 UTC+8

Pentagon Tests Rival AI Models in Race to Replace Anthropic

The Pentagon is testing rival AI models with 25 departmental “power users” as it seeks alternatives to Anthropic’s Claude, according to a senior defense official; the RSS snippet does not disclose the candidate model list, evaluation criteria, or deployment timeline.

Why it matters: Bloomberg sourcing plus Pentagon testing rivals to Anthropic clears HKR-H/K/R. Candidate models, contract size, and timeline are not disclosed, so it sits just above the featured threshold.

Read the original ↗Export Markdown

Take

I would not read this as Anthropic losing the Pentagon yet; 25 “power users” is a procurement probe, not a model bake-off. The snippet says the department wants alternatives to Claude, but gives no candidate list, scoring rubric, deployment date, or task mix. We do not know if users tested office drafting, intel analysis, code, classified RAG, or policy review.

The sharper signal is that Claude is named as the incumbent to beat. Anthropic has sold hard into the safety-and-governance lane, where defense buyers like auditability and refusal behavior. A rival test lets the Pentagon avoid vendor lock-in and pressure pricing or terms. If the list includes OpenAI, Google, Meta, or Palantir-wrapped models, the read changes fast. With only 25 testers disclosed, Bloomberg’s frame is ahead of the evidence.

Source: Bloomberg Technology · bloomberg.com
Published: May 22, 2026, 01:38 UTC+8 · May 22, 2026