The Lyceum: Defense Tech Daily — Mar 13, 2026

Iran's new supreme leader may be issuing war orders from a hospital bed, Canada just announced it's building an Arctic military it doesn't need American permission to use, and the Pentagon quietly admitted it has no standardized way to check whether the AI it's already using in combat actually works as advertised. The theme of the day: institutions are making enormous bets at speed, and the verification is running months behind the deployment.

The Iranian regime released Mojtaba Khamenei's first public message Thursday — a hardline declaration that the Strait of Hormuz stays closed, U.S. bases should shut down or face attack, and Iran will avenge its dead. The problem: multiple Western outlets, citing sources inside Tehran, report the new supreme leader may be in a coma, severely injured in the same February 28 strike that killed his father. The Sun reported he lost a leg and has critical organ damage, though Iran's internet blackout makes verification impossible.

This matters less as a health story than as a power story. A regime issuing statements in the name of a leader who can't speak for himself is, by definition, a regime running on institutional momentum rather than command authority. And the statement itself directly contradicted Iranian President Masoud Pezeshkian, who floated ceasefire conditions just the day before. If Pezeshkian starts making solo decisions without attributing them to Khamenei, the war's off-ramp negotiations shift dramatically.

The man now controlling Iran's nuclear program and Strait of Hormuz strategy was already sanctioned by the U.S. Treasury in 2019 for advancing his father's "destabilizing regional ambitions." Trump called him "a lightweight." The IRGC treated that insult as an endorsement. He is more connected to Iran's security establishment than his father ever was — and may be running the war from an ICU.

The most uncomfortable headline of the day comes from inside the Israeli security establishment. Israeli security sources told reporters Thursday that the February 28 operation had no clear plan for regime change once the bombs started falling.

The battlefield numbers look impressive on paper — Iranian ballistic missile attacks are down 90 percent since the Feb. 28 operation, and drone attacks are down 83 percent since the Feb. 28 operation. But those numbers measure military degradation, not political collapse. Despite the deaths of Supreme Leader Ali Khamenei and other senior figures, the Islamic Republic is holding together with no major leadership defections. Instead of mass protests, domestic anger has shifted toward existential solidarity.

The "Venezuela solution" — replacing the top leader with a cooperative subordinate — has already failed. Iran installed a harder-line figure than the one the U.S. killed. Planners reportedly used a "cyber-first" posture in the opening phase of Operation Epic Fury, blinding air defenses before kinetic strikes, which suggests Washington expected non-kinetic tools to create windows that conventional force alone couldn't. The question now isn't whether the military campaign is working. It's what "winning" even looks like.

A defense official told MIT Technology Review that a list of possible targets can now be fed into a classified generative AI system, which humans then ask to rank priorities — accounting for factors like current aircraft locations. The official confirmed this is already reducing targeting time, though declined to say by how much.

The technical detail that matters: generative AI — the same large language model technology behind ChatGPT — is being layered as a conversational front-end on top of Project Maven, the Pentagon's legacy computer vision targeting tool running since 2017. You don't just get dots on a map anymore; you get a system that explains which dot to hit first, and why. The old interface forced users to directly inspect raw data. The new one is easier to access but harder to verify.

This landed the same week a group of 120-plus House Democrats sent a letter to Defense Secretary Pete Hegseth on March 12 asking whether AI — specifically Maven — selected the Shajareh Tayyebeh school as a target. Preliminary reporting found outdated targeting data was partly responsible. The lawmakers' deadline for answers is March 20. The most direct congressional challenge to AI-enabled targeting ever put to a defense secretary, and it hinges on a question this newsletter keeps returning to: if the data going in is stale, does it matter how fast the AI processes it?

The Pentagon and Defense Innovation Unit handed Anduril a contract to build extra-large autonomous underwater vehicles — uncrewed submarines designed for long-range missions where crewed subs can't go or shouldn't risk going. Anduril's Dive-XL prototype demonstrated the endurance needed to win the program, and the contract requires an extended ocean demonstration within four months of award — a timeline that underscores how urgently the Navy wants to push missions underwater into robotic platforms.

This fits a week-long pattern: Japan announced a 100 billion yen ($675 million) drone "Shield" for coastal defense, the Navy is preparing to integrate autonomous surface ships into carrier strike groups, and L3Harris and Shield AI reported a breakthrough test where unmanned aircraft autonomously detected and reacted to electronic warfare threats without human input. Across every domain — undersea, surface, air, electromagnetic spectrum — the answer to "who goes first into danger?" is increasingly "a robot."

A supreme leader who may be governing from a coma, a country building Arctic air bases because its best friend keeps threatening to annex it, and a Pentagon that deployed AI to pick targets before building a system to check whether the AI works. The 46-to-1 ratio of Pentagon AI users to trained users is the kind of number that sounds like a rounding error until it's in the footnote of an incident report. Stay sharp.

If someone you know should be reading this, send it their way.