The Lyceum: Defense Tech Daily — Mar 19, 2026

The Pentagon is simultaneously trying to build the most classified AI ever conceived and rip out the one it already has — while burning through munitions faster than American factories can make them. The war with Iran just hit the global economy's most vulnerable pressure point, and a supplemental funding request to Congress is $200 billion.

Your heating bill just became a war story. Ras Laffan Industrial City in Qatar sustained damage from Iranian missiles on Wednesday night — the single facility responsible for roughly a fifth of the world's liquefied natural gas. Qatar called the damage "extensive." The plant was already offline since early March; this strike appears structural, not cosmetic. One analyst told gCaptain it's "hard to see Qataris coming back to the market before the middle of the year." Citigroup flagged a scenario where Brent crude averages $130 a barrel through summer if broader energy infrastructure keeps getting hit while the Strait of Hormuz remains contested, according to CNBC.

Here's the tactical detail that matters strategically: four Iranian missiles were intercepted over Ras Laffan. One got through. A Patriot interceptor costs roughly $3–4 million; the attacking missile costs a fraction of that. When defenders spend tens of millions stopping a volley and a single warhead still damages a facility that supplies a large share of global LNG, it raises questions about the economics of missile defense. Iran appears to be shifting from counter-military to counter-economy targeting — an escalation that raises concerns about whether supply chains can absorb repeated infrastructure shocks. Qatar expelled Iranian military attachés hours later, per Al Jazeera.

Here's a number that reframes the entire war: $200 billion. That's what the Pentagon asked the White House to request from Congress to fund the Iran campaign, according to the Washington Post. The striking part isn't the size — it's the purpose. This is largely about urgently ramping production of precision munitions after a three-week air campaign burned through stockpiles built for deterrence, not sustained high-intensity war. Officials told the Boston Globe that early costs exceeded $11 billion in the first week alone. The Detroit News reports the request also covers AI targeting systems, drone countermeasures, and lower-cost strike options like the near-hypersonic missiles we've been tracking. Deputy Defense Secretary Steven Feinberg is reported to be leading efforts to accelerate production, but passage of a supplemental appropriations bill on the Senate floor would require invoking cloture (60 votes). If lawmakers fail to secure cloture, operational tempo — how fast and hard the U.S. can strike — risks being dictated by factory output, not strategy.

The Anthropic lawsuit is moving faster than most people realize. The DOJ filed papers arguing the Pentagon's ban on Anthropic's Claude AI is "lawful and reasonable," framing the dispute as contract negotiations, not retaliation. The government's core claim: if an AI company can embed operational red lines into government contracts, "an AI provider might gain influence over how the DoD conducts operations and which missions it chooses." The filing goes further, raising the specter that Anthropic could "attempt to disable its technology or preemptively alter the behavior of its model either before or during ongoing warfighting operations," according to Engadget. Dawn reports the government has labeled Anthropic an "unacceptable risk" — language normally reserved for adversary vendors like Huawei. If this argument holds, it sets the terms for every AI deal the Pentagon signs going forward: no vendor gets to decide which missions their code supports.

Ukraine's fight for survival has moved into the boardrooms of Boston, Austin, and San Francisco. Sixteen Ukrainian defense-tech startups just wrapped a "USA Investment Roadshow" organized by Brave1, the government-backed incubator. These companies build GPS-denied navigation systems, resilient battlefield communications, and drone autonomy software — all tested under fire. The timing isn't accidental: Swarmer, a Ukrainian-founded drone-swarm AI company, IPO'd on Nasdaq this week and surged over 500% on its first trading day per Crunchbase. Investors are buying combat provenance, not pitch decks. If a top-tier U.S. VC leads a round for one of these roadshow companies, it'll be the clearest signal yet that Ukraine's tech base is being woven permanently into Western supply chains.

Right now, military AI reads classified documents and answers questions — like a human analyst with a security clearance. The Pentagon wants to go further: let AI companies train models on classified data inside secure, air-gapped government facilities, according to MIT Technology Review. The difference matters enormously. A model that absorbs thousands of intelligence reports restructures its own internal weights around them. That's not a chatbot with access — it's a system with secrets baked into its architecture. The biggest risk, per the same reporting: classified information could be "resurfaced to anyone using the model," potentially leaking an operative's identity to a department that shouldn't have it. Meanwhile, Ukraine is running the opposite experiment — sharing sanitized battlefield data with allies to train drone AI without exposing raw files. Two models of the same idea, two very different risk profiles.

A single warhead got through multiple interceptors and damaged a facility that supplies roughly a fifth of the world's LNG. A DOJ lawyer arguing that an AI company's safety culture is itself a national security threat. Sixteen Ukrainian startups pitching VCs in San Francisco while their countrymen field the drones those VCs want to fund. The Pentagon is asking Congress for $200 billion and simultaneously trying to figure out whether the AI it's buying can be trusted not to leak the secrets it's trained on — which is the kind of problem you get when you build the plane, the engine, and the runway all at the same time, in a war zone, while suing your mechanic.

Stay sharp.

If someone you know needs to understand what's actually happening — not the headline, the machinery underneath — forward this their way.

The FTC is treating "unfair" AI as a punishable business practice — turning ethics debates into real legal exposure for companies building these systems. Read → FTC Draws a Line: "Unfair" AI Is Now an Enforcement Target Under Section 5