AMZN
GOOG
MSFT
NVDA
The Pentagon’s AI shopping cart got stricter
The Defense Department just made a very 2026 move: it wants more AI, but with fewer trust issues. On Friday, the Pentagon’s tech chief said Anthropic’s Claude models are still a supply-chain risk, and some departments may phase them out over the coming months.
That’s not exactly a ringing endorsement. It’s more like, “We like the product, just not enough to let it anywhere near the sensitive stuff.”
Not a total ban, more of a government-size asterisk
Emil Michael said there can be exceptions when integration gets messy, but the broader message was clear: Claude’s status is still a national security problem. He even framed Mythos as a separate issue — a broader, government-wide concern tied to cyber defense and network hardening.
In plain English, the Pentagon is worried that the smarter the model gets, the more dangerous it could be if it’s wired into the wrong systems.
The twist: the DoD still wants everyone in the room
At the same time, the Defense Department announced agreements with seven AI companies, including:
- Alphabet’s Google
- OpenAI
- Nvidia
- Microsoft
- Amazon
- SpaceX
- Reflection, a newer startup
So the government isn’t backing away from AI. It’s diversifying like a cautious portfolio manager who finally read the fine print. The message seems to be: don’t bet the whole national-security stack on one vendor, and don’t expect the guardrails to be non-negotiable.
Why investors should care
This is good news for the broader AI infrastructure race, because the Pentagon is still spending. But it’s also a warning label for companies trying to sell AI into the most sensitive corners of government: security review, integration hurdles and policy constraints can slow adoption fast.
Big picture: the DoD wants AI in the building — it just wants the fire exits, locks and security cameras first.