Pentagon Picks 8 AI Vendors for Classified Networks. Anthropic Is Not One of Them
The Pentagon just cleared eight tech companies to deploy AI on its IL6 and IL7 classified networks. AWS, Google, Microsoft, NVIDIA, OpenAI, SpaceX, Reflection, and Oracle. The list is interesting for who's on it. The list is more interesting for who's not.
Anthropic is the absence everyone is reading. Claude is already running on classified networks via Palantir's Maven toolkit, so technical fit was never the question. The question was terms. The DoD wanted unrestricted use. Anthropic insisted on guardrails against domestic mass surveillance and autonomous weapons. Pentagon CTO Emil Michael spelled it out: it's irresponsible to be reliant on any one partner, and that one partner didn't really want to work with us in the way we wanted to work with them. Translation: Anthropic walked.
Reflection AI being on the list is the other read worth noting. Reflection is the NVIDIA-backed agent startup founded last year by ex-DeepMind researchers. It just put itself in the same bracket as OpenAI and Microsoft at IL6 and IL7 β the highest classification tiers in the U.S. government. A twelve-month-old company sharing a row with AWS at the Pentagon is what the sovereign AI bet looks like when it actually lands.
The wider story is that the U.S. military is now actively diversifying away from any single AI vendor. Eight names is not a deal. Eight names is a portfolio. Read it as the Pentagon telegraphing that frontier AI is now a regulated infrastructure category, with multiple approved suppliers and one notable gatekeeper that walked the door. Combined with the GPT-5.5-Cyber restricted-access launch yesterday and the UK AISI sabotage research the week before, the access gradient on frontier capability is now visible in a way it wasn't six months ago.
https://breakingdefense.com/2026/05/pentagon-clears-7-tech-firms-to-deploy-their-ai-on-its-classified-networks/
← Back to all articles
Anthropic is the absence everyone is reading. Claude is already running on classified networks via Palantir's Maven toolkit, so technical fit was never the question. The question was terms. The DoD wanted unrestricted use. Anthropic insisted on guardrails against domestic mass surveillance and autonomous weapons. Pentagon CTO Emil Michael spelled it out: it's irresponsible to be reliant on any one partner, and that one partner didn't really want to work with us in the way we wanted to work with them. Translation: Anthropic walked.
Reflection AI being on the list is the other read worth noting. Reflection is the NVIDIA-backed agent startup founded last year by ex-DeepMind researchers. It just put itself in the same bracket as OpenAI and Microsoft at IL6 and IL7 β the highest classification tiers in the U.S. government. A twelve-month-old company sharing a row with AWS at the Pentagon is what the sovereign AI bet looks like when it actually lands.
The wider story is that the U.S. military is now actively diversifying away from any single AI vendor. Eight names is not a deal. Eight names is a portfolio. Read it as the Pentagon telegraphing that frontier AI is now a regulated infrastructure category, with multiple approved suppliers and one notable gatekeeper that walked the door. Combined with the GPT-5.5-Cyber restricted-access launch yesterday and the UK AISI sabotage research the week before, the access gradient on frontier capability is now visible in a way it wasn't six months ago.
https://breakingdefense.com/2026/05/pentagon-clears-7-tech-firms-to-deploy-their-ai-on-its-classified-networks/
Comments