- The Department of Defense signed classified AI agreements with eight companies on May 1, 2026: OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX and Reflection AI.
- Anthropic was deliberately excluded after the Pentagon designated the company a “supply chain risk” in early March, a label normally reserved for foreign-adversary contractors.
- The dispute traces back to Anthropic’s refusal to remove guardrails blocking Claude from being used for autonomous weapons or domestic mass surveillance.
- Anthropic has filed two lawsuits and won a preliminary injunction in California, but lost an appellate stay in DC; the supply-chain designation is still being contested.
Did the Pentagon just pick favorites in the AI race? On May 1, 2026, the Department of Defense quietly cleared eight of the largest AI vendors in the world for work on classified IL6 and IL7 networks. The list reads like a who’s-who of frontier AI: OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX and Reflection AI. One name is conspicuously missing. Anthropic, the company that builds Claude and that posted a $30 billion ARR run-rate this quarter, was cut out entirely.
What Actually Happened on May 1
Eight winners, one omission
The agreements expand DoD use of generative AI into classified environments for analysis, logistics, code generation and large-scale data processing. According to DefenseScoop and Federal News Network, the contracts let each vendor deploy models inside the department’s most sensitive Impact Level 6 and Impact Level 7 enclaves — the layers used for Secret and Top Secret workloads. CNN Business confirmed Anthropic’s omission was deliberate, not an oversight, and traced it back to the “supply chain risk” tag the Pentagon attached to the company in March.
A label normally aimed at adversaries
National security analysts who spoke to Al Jazeera and CBS News described the supply-chain risk designation as “highly unusual” when applied to a U.S. company. The label is historically reserved for vendors tied to foreign adversaries that could plausibly sabotage U.S. systems. Using it against an American AI lab valued north of $180 billion is essentially unprecedented — and that is exactly why the May 1 announcement landed with the force it did.
Trend Insight — The Pentagon is no longer just buying AI; it is shaping the policy posture of every lab that wants to sell to it. The signal sent on May 1 is that vendors who keep hard guardrails on weaponization may be priced out of classified work, regardless of model quality.
Why Anthropic Was Cut
The “all lawful purposes” demand
According to NPR and Fortune, Defense Secretary Pete Hegseth pushed Anthropic to accept terms that would allow Claude to be used for “all lawful purposes” inside DoD systems. That phrase is critical, because the Pentagon’s interpretation explicitly included two lines Anthropic refused to cross: fully autonomous lethal weapons and domestic mass surveillance. Anthropic’s position, repeated in public filings, is that current frontier models are not yet reliable enough to authorize lethal action without a human in the loop, and that exposing the U.S. population to model-driven surveillance falls outside its acceptable-use policy.
From negotiation to blacklist
When Anthropic held the line, Hegseth escalated. In early March the Pentagon formally designated the company a “supply chain risk,” a procurement-killing label that effectively bars federal agencies from buying Claude. Anthropic responded with a one-two legal punch on March 9: a federal complaint in the Northern District of California, and a parallel petition before the DC Circuit. The company argues the designation is unlawful and infringes its First Amendment and due-process rights.
Trend Insight — This is the first time a frontier AI lab has tested whether usage-policy enforcement is protected speech. The case will set a template for how every other lab negotiates with a customer that wants its safeguards stripped.
Court Rulings and the Road Ahead
A split scoreboard
CNBC reported that a federal judge in San Francisco granted Anthropic a preliminary injunction last month, blocking the administration from enforcing a separate Claude ban while the underlying litigation continues. On April 8, however, the DC Circuit denied Anthropic’s request for a stay in its lawsuit against the DoD, leaving the supply-chain risk tag in force pending merits review. The May 1 contract awards were structured to operate cleanly inside that legal gap.
What it means for the AI vendor market
For OpenAI, Google and Microsoft, the deal is leverage. Federal classified work has historically anchored long-tail enterprise revenue, and being inside the IL6/IL7 perimeter creates a moat competitors can’t cross without years of accreditation. For Anthropic, the financial impact is more reputational than topline — classified DoD spend is a small fraction of its $30B ARR — but the long-term risk is that allied governments and Fortune 500 procurement teams begin treating “Pentagon-cleared” as a baseline trust signal.
Trend Insight — Watch the next earnings cycle. If enterprise customers start asking “are you Pentagon-cleared?” during AI vendor diligence, the supply-chain risk tag transforms from a regulatory line item into a market-shaping economic penalty.
Related
- Anthropic’s CEO Just Issued a 6-Month Warning
- Goldman, JPMorgan and the Wall Street AI Land Grab
- OpenAI Chair’s Side Project Is Now Worth $15.8B
- The Clipboard Attack Hidden Inside an AI Suggestion
Sources
- CNN Business — Pentagon strikes deals with 8 Big Tech companies after shunning Anthropic
- DefenseScoop — DOD expands classified AI work with 8 companies, excluding Anthropic
- Al Jazeera — Anthropic sues Trump administration to undo supply chain risk tag
- CNBC — Anthropic loses appeals court bid to temporarily block Pentagon blacklisting
AI Biz Insider · AI Trends EN · aibizinsider.com
댓글 남기기