The Pentagon Is Spending $14 Billion on AI Weapons. Most of the World Has No Idea.
The US military's record $14.2 billion AI weapons budget includes armed humanoid robots and autonomous drones — invisible to 6 of 7 global regions.

The Pentagon just created its first-ever budget line for AI weapons: $13.4 billion. Add the $151 billion reconciliation act for drones, autonomous systems, and military AI, and the US is spending more on robot warfare than most countries spend on their entire militaries.
Almost nobody outside America knows.
The Global Attention Index scores this at 7 out of 10 — covered in one region, invisible to the other six. About 7.3 billion people live in countries where this arms race isn't making headlines.
The Numbers Behind the Machine
This is the first time the Department of Defense gave AI and autonomy their own line item. It's roughly a 7x increase in a single budget cycle.
But that's just the formal budget. The One Big Beautiful Bill Act added $151 billion in mandatory defense spending. Defense One obtained the allocation plan: $1.4 billion to expand the drone industrial base. $500 million for a new Defense Autonomous Warfare Group. $650 million for "multi-domain collaborative autonomy" — sea, air, and land drones coordinating missions without humans.
The Navy alone gets $1.5 billion for small surface drones, $2.1 billion for autonomous boats like the Sea Hunter, and $1.3 billion for underwater drones. The Army gets $74 million for its Autonomous Ground Fighting Vehicle. And $145 million goes to kamikaze drones that pick their own targets.
The Robot That Carries a Gun
Then there's Phantom.
Foundation, a US startup, built what TIME calls "the world's first humanoid robot for defense." The Phantom MK-1 stands in jet-black steel with a tinted visor. It carries revolvers, shotguns, and M-16 replicas. Co-founder Mike LeBlanc — a 14-year Marine veteran — says it can wield "any kind of weapon that a human can."
Foundation holds $24 million in contracts with the Army, Navy, and Air Force. Two Phantoms went to Ukraine in February for frontline recon. The company's testing door-breaching with the Marine Corps and talking to Homeland Security about border patrol.
They want to build 50,000 armed humanoids by 2027.
"We think there's a moral imperative to put these robots into war instead of soldiers," LeBlanc told TIME.
Already Happening in Ukraine
This isn't theoretical. AI-powered drones in Ukraine already select targets and fire on their own. Russian radio jamming broke remote control, so manufacturers built AI that lets drones lock on and strike alone.
Ukraine's Defense Minister Mykhailo Fedorov opened millions of drone strike videos as training data for AI targeting. The goal: autonomous weapons that can't be jammed.
The Replicator initiative — the Pentagon's program for "thousands of autonomous systems" — promised those thousands by summer 2025. It delivered hundreds. But $500 million now goes to a Defense Autonomous Warfare Group to fix the bottleneck. Phase two is accelerating.
The Fight Over Red Lines
This buildup is splitting the AI industry. Anthropic refused to let its Claude model power autonomous kill systems or domestic mass surveillance. The Pentagon called Anthropic a "supply chain risk." Anthropic's fighting that in court.
OpenAI stepped in, grabbing Pentagon contracts. The Electronic Frontier Foundation warned OpenAI's deal has loopholes that "won't stop AI-powered surveillance."
In Geneva, talks on regulating lethal autonomous weapons have stalled. A UN resolution in November 2025 called for a binding agreement by 2026. It passed 156-5. The five who voted no: the US and Russia among them.
Why Six Regions Can't See This
American defense publications and tech outlets cover this closely. Everywhere else? Silence. Europe, the Middle East, Asia-Pacific, South Asia, Africa, Latin America — nothing.
It's not that it doesn't affect them. The US is building autonomous naval systems for the Pacific. Japan just deployed long-range missiles. AI drone tech tested in Ukraine will shape every future war. And the regulation talks collapsed because the nations building these weapons blocked them.
Jennifer Kavanagh of Defense Priorities told TIME: "The appeal of automating things and having humans out of the loop is extremely high. The lack of transparency between the two sides of any conflict creates additional concerns."
What Happens Next
The next 18 months decide whether autonomous weapons get normalized or constrained. The $151 billion must be spent by September 2029. Foundation wants tens of thousands of armed humanoids deployed. Ukraine's open-source battlefield data is training the next wave of targeting AI.
Geneva talks resume later this year. But the US and Russia both voted against binding rules, and autonomous weapons are already firing in combat. The window for limits is shrinking.
Seven billion people don't know this race is underway. The machines are being built anyway.
This story was identified by the Albis Global Attention Index — measuring which stories the world isn't seeing. Explore today's blind spots →
Sources & Verification
Based on 5 sources from 4 regions
- Defense OneNorth America
- TIMENorth America
- ReutersInternational
- The GuardianEurope
- Usanas FoundationSouth Asia
Keep Reading
300 Google Employees Just Drew a Line the Pentagon Can't Cross
AI workers at Google, OpenAI, and Anthropic are refusing to build weapons and surveillance tools. The deadline is tonight.
The Pentagon gave Anthropic a Friday deadline. This is what happens when AI safety meets national security.
The only frontier AI with classified Defense Department access just refused to remove usage restrictions. The Pentagon threatened to invoke the Defense Production Act. Friday is the deadline.
China Just Set Its Lowest Growth Target Since 1991 — and Most of the World Missed It
China cut its GDP target to 4.5-5% at the Two Sessions, the lowest since 1991. Here's why 6 billion people should care.
Explore Perspectives
Get this delivered free every morning
The daily briefing with perspectives from 7 regions — straight to your inbox.