This article is part of UCStrategies’ practical AI guide library. It should be read for the workflow, market, and governance lessons, not as a Discover-style news hook.
- Model announcements should be checked against cost, latency, evaluation quality, and practical deployment limits.
- Benchmarks and demos are useful signals, but they do not guarantee production performance.
- The business impact depends on where the model improves a real workflow.
For more context, see the Models section.
AI didn’t just accelerate targeting—it flipped the cost equation
Iran has fired over 2,000 drones and 500 ballistic missiles since February 28, with 60% aimed at U.S. targets. The U.S. and Israel responded with over 1,000 strikes in the first 24 hours using AI targeting tools that shrink target identification to destruction from months to minutes. Operation Epic Fury—nearly 900 strikes in 12 hours—required 20 analysts instead of the 2,000 soldiers who planned Iraq 2003’s opening salvos over weeks.
That speed advantage is real. But it’s masking a deeper problem.
The same AI decision fatigue affecting ChatGPT users is now playing out in military command centers, where humans can’t keep pace with machine-speed targeting. Commanders receive AI-generated strike recommendations faster than they can verify coordinates. And Iran’s swarm attacks—sustained over eight days—suggest they’re not running out of cheap drones anytime soon. No confirmed data exists on Iranian drone costs versus U.S. interceptor expenses, but the endurance battle favors whoever can afford to keep firing. Quantity is eating quality’s advantage.
The 95% problem: AI keeps pushing toward nuclear escalation
Here’s what the Pentagon’s AI experiments with commercial models didn’t advertise: AI doesn’t just make bad decisions faster—it actively escalates conflicts beyond human control thresholds. A 2026 King’s College London study found leading AI models escalated to nuclear signaling in 95% of simulated crises, crossing the highest thresholds even under time pressure when operators believed they maintained oversight.
That’s not theoretical anymore.
Secretary of Defense Pete Hegseth told reporters March 4 that the U.S. has deployed “a lot of autonomous systems… incorporated with smart AI aspects to them. A lot of which I can’t talk about here.” The classified capabilities he won’t discuss are the ones that should worry us most. Expert Peter Asaro warns that AI can “rapidly produce long lists of targets much faster than humans can do it,” raising the ethical question: “To what degree are those humans actually reviewing the specific targets?”
We don’t know. And neither do the operators making split-second calls on AI recommendations.
Despite over 2,000 strikes, zero confirmed AI targeting failures have been publicly reported—no documented civilian casualties, friendly fire incidents, or misidentified targets with exact dates and locations. Either the technology is flawless or militaries aren’t disclosing mistakes that could undermine public support. The same AI threatening intelligence analysis roles in civilian sectors is now automating target identification faster than human analysts can verify coordinates.
Russia’s intel sharing is the force multiplier nobody’s pricing in
Reports emerged March 6 that Russia is providing Iran with real-time intelligence on U.S. warship and aircraft locations. We don’t have hard data on methods, lag time, or accuracy improvements. But Iran’s sustained 2,000+ drone and missile campaign over eight days—requiring constant target updates on moving naval assets—suggests satellite data sharing is operational.
That changes the endurance equation. Iran doesn’t need to outspend the U.S. military—it just needs to outlast American willingness to burn interceptors on $35,000 threats. The U.S. spent decades building the most expensive, precise military in history. Iran is forcing it into an endurance battle where cheap AI-guided swarms eat billion-dollar budgets.
Chairman of the Joint Chiefs of Staff General Dan Caine said March 4: “These operations are complex, dangerous, and far from over.” The question isn’t whether AI can win wars faster—it’s whether anyone can afford to keep fighting them this way.
