How Project Maven taught the military to love AI

The real story isn't that AI assisted a military strike — it's that the targeting workflow compressed to a scale (1,000+ targets in 24 hours) that would have been logistically impossible without automation, which means the human-in-the-loop question is no longer theoretical. The bottleneck has shifted from capability to accountability.

This lands directly against the Anthropic coverage from mid-April. Both the Verge piece on Claude Mythos Preview and the TechCrunch report on Anthropic's thawing relationship with the Trump administration show AI labs actively repositioning toward government contracts. Maven Smart's operational debut clarifies what that repositioning is actually for: not general federal IT, but systems that sit inside lethal decision chains. Labs courting DoD access now have a concrete, public benchmark to measure themselves against, and a much harder set of questions to answer about where their models fit in that workflow.

Watch whether Congress moves to attach oversight requirements to Maven Smart's next procurement cycle within the next six months. If it doesn't, that signals the operational use case has outrun the policy debate entirely, and the accountability gap becomes structural rather than temporary.