Pentagon signs seven AI deals for classified work

This is less a product launch than a procurement signal: the Pentagon wants a multi-vendor AI stack it can push into operational workflows fast. The bigger question is not whether the tech works, but which safety constraints survive when defense buyers want broad “lawful use.”

• –The deals cover OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, Reflection, and SpaceX, with Anthropic notably absent
• –Deployment into IL6 and IL7 environments means these systems are moving into highly sensitive classified infrastructure, not just pilot programs
• –The Pentagon is explicitly trying to avoid vendor lock-in, which suggests AI procurement is becoming a strategic infrastructure layer
• –The unresolved risk is human oversight: autonomous use, target assistance, and domestic surveillance concerns are still the core pressure points
• –For vendors, defense revenue is attractive, but the ethical and reputational cost is rising fast