The US military hit more than 1,000 targets in the first 24 hours of its assault on Iran, nearly double the scale of the 2003 “shock and awe” campaign against Iraq. According to The Verge AI, the engine behind that pace is the Maven Smart System, the subject of a new book by journalist Katrina Manson titled Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare. The Verge AI reports that Maven now compresses targeting work that once took hours into seconds, and recently got picked up by NATO.
The numbers are the story. An official told Manson that Maven helped the US go from striking under 100 targets a day to 1,000, and with large language models layered in, up to 5,000 a day. That’s a 50x jump in operational tempo driven almost entirely by software.
How Maven actually works
Manson describes Maven as computer vision wrapped in a workflow tool. It pulls from satellite imagery, radar, social media, and dozens of other feeds, then:
- Identifies entities on the battlefield
- Pairs targets with available weapons
- Walks operators through the targeting cycle with a few clicks
This is what military planners call the “kill chain.” Maven shortens it. The acceleration is the product.
The contractor reshuffle nobody talks about
Maven started in 2017 as a Pentagon experiment in applying computer vision to drone footage. Google was the original contractor. Employee protests in 2018 pushed the company out, and Google publicly framed the work as “non-offensive.” Manson’s reporting tells a different story. One source in the book put it bluntly: “Yeah, of course, it’s not like we’re doing it for kicks. The goal of the intel is to take out high-value targets.”
When Google walked, the vacuum filled fast. Palantir took the lead integrator seat. Microsoft and Amazon stepped up on algorithms and compute. Anthropic’s tech got pulled in too. What looked like a values win for Google employees turned out to be a contract handoff.
Why the Iran strike matters for the industry
One of the 1,000 first-day targets was a girls’ school. More than 150 people died, mostly children. The site had once been part of an Iranian naval base, but it was listed online as a school and the playgrounds were visible on satellite imagery. Coverage initially blamed possible Claude hallucinations. Technology historian Kevin Baker pushed back in The Guardian: “A chatbot did not kill those children. People failed to update a database, and other people built a system fast enough to make that failure lethal.”
That framing is the real industry signal. Speed is now a weapon, and speed magnifies the cost of bad data.
What this means for AI vendors and operators
Three shifts worth tracking:
- Defense is no longer optional revenue. Palantir’s market cap and the NATO Maven purchase show where the spend is heading. Vendors who sat out a few years ago are quietly back in.
- Data hygiene becomes a safety system. When the kill chain runs in seconds, stale records aren’t admin debt. They’re casualties.
- The next wave is autonomous. Manson uncovers Pentagon programs for fully autonomous weapons, including an explosive-laden drone Jet Ski. Human-in-the-loop is the current marketing line. It’s not the roadmap.
Practical takeaways
- If you build infrastructure or models, assume defense use is on the table whether you ship there directly or not. Decide your policy before the contract shows up.
- If you work in AI safety or evals, the Maven case is the new reference point. The failure mode wasn’t a model. It was a workflow that outran its own ground truth.
- If you advise enterprises, watch the Palantir playbook. Tight integration plus government contracts is becoming the moat that pure-model companies can’t match.
Manson’s book lands as Western militaries normalize AI-driven targeting and rivals race to match it. The next 18 months will decide whether “human oversight” stays a real constraint or becomes a slogan stapled onto fully automated systems. Full reporting is available at the original source.