[Key Facts] The Pentagon has stopped using AI tools from Anthropic after placing the company on an internal restricted-vendor list, according to a BBC exclusive. Internal Defense Department documents show the military is pivoting to OpenAI and Google alternatives.

[What We Know]

  • Internal ban: The Department’s Digital and AI Office has directed subordinate units to “immediately cease all calls to Anthropic products, including commercial APIs and open-weight models.”
  • Alternatives under evaluation: The Pentagon is assessing OpenAI’s GPT family, Google’s Gemini, and the open-source Llama 4; some intelligence-analysis workloads have already migrated to GPT-5 Turbo.
  • Underlying reason: Sources say the restriction stems from a mismatch between Anthropic’s “safety alignment” guarantees to enterprise customers and the Pentagon’s data-flow requirements. No official explanation has been published.
  • Industry impact: Analysts estimate Anthropic’s near-term government-contract revenue will fall roughly 15%, putting pressure on US AI-startup share in federal procurement.

[Wider perspective] First — The Pentagon’s rapid supplier switch signals that US military procurement is prioritizing “model interpretability and data sovereignty” over “model scale and raw performance.” Frontier-model vendors will need to make firmer commitments on architectural transparency and auditable reasoning to remain in federal supply chains.

Second — Anthropic’s setback comes amid the most intense phase yet of the OpenAI–Anthropic contest for enterprise API share. Loss of Defense Department contracts may further pressure Anthropic’s valuation trajectory, while accelerating concentration of government market share in OpenAI and Google — a clear AI-head-consolidation trend.

Third — This episode is part of Washington’s escalating narrative on AI “safety risk.” Following this year’s congressional pushes on the Frontier AI Accountability Act and the Defense AI Auditability Act, the Pentagon’s “blacklist” move is in lockstep with legislative momentum, marking a shift from voluntary frameworks to mandatory AI governance.

[Viewpoints Compared]

  • Pentagon spokesperson: Declined to comment on specific vendors, saying only that the Department “regularly reviews the safety compliance of all AI partners.”
  • Anthropic spokesperson: Did not address the blacklist directly, stating only that “the company continues dialogue with federal agencies.”
  • US Senate Armed Services Committee member: Called on the Pentagon to publish its “blacklist” criteria, warning that opacity could undermine US AI industry competitiveness.
  • Stanford AI Policy expert: Said the move shows the Pentagon can “de-risk rapidly,” but lacks an open appeal mechanism.
  • OpenAI government affairs lead: No direct comment, but stressed the company’s “long-term stable service commitment to defense customers.”

Editor: GoodInfo Global News Team