Four Days into a New War, Four Experts Weigh In
Four days into American airstrikes on Iran, ChinaTalk host Jordan Schneider convened an emergency panel featuring Emmy Probasco from the Center for Security and Emerging Technology (CSET), Henry Farrell of Johns Hopkins, University of Pennsylvania professor Mike Horowitz, and Bryan Clark of the Hudson Institute. The resulting conversation is a rare real-time policy debate that spans kinetic warfare, weapons stockpile math, nuclear proliferation, and the surprisingly bureaucratic reality of artificial intelligence in combat operations.
The panel wastes no time on pleasantries. Horowitz opens with a striking technical detail about the weapons being used.
What's notable is that the United States used a system called the LUCAS, which is America's first precise mass system. It costs less than $100,000 and can travel a couple thousand kilometers. You can shoot it down, but you have to try.
The irony is not lost on the panelists: LUCAS is reverse-engineered from Iran's Shahed 136, which itself borrowed from West German technology from the 1980s. As Horowitz quips, "what goes around comes around." The episode title references "precise mass," a military concept describing the ability to deliver large quantities of precision-guided munitions simultaneously. Schneider frames this capability against Robert Pape's influential thesis in "Bombing to Win" -- that strategic bombing historically fails to coerce adversaries into surrender. The difference in 2026, Schneider argues, is that precision strikes can now eliminate an entire leadership class from thousands of miles away.
The Missing Piece: Who Picks Up After Decapitation?
Clark raises what may be the panel's sharpest strategic objection. Air power can kill leaders, but it cannot build governments.
Any competent autocrat in the 21st century will eliminate potential competition. It's not like when the British faced the American Revolution -- we had people who could take charge, and they didn't bother assassinating them in advance.
Schneider pushes harder on the timing. If regime change was the goal, the window was during the 2022 protests, not after thirty thousand of the most politically active Iranians were already dead. Horowitz concedes the point but notes that assembling assets for a sustained campaign takes months -- you cannot simply snap your fingers and deploy at scale.
The panel collectively struggles to identify any coherent American theory of victory. Horowitz is blunt about the incoherence.
We've never seen the United States attempt a military operation of this scale with such incoherent goals. Sometimes it sounds like regime change; sometimes it's about "eliminating the threat" -- whatever that means.
He draws a dark analogy to "The Godfather" -- the Trump administration systematically working through a list of adversaries: Iran, then Cuba, then potentially others. Clark calls it "taking care of the family business." The gallows humor underscores a genuine worry: no one on the panel can articulate what success looks like beyond the initial bombardment.
Iran Is Still Shooting Back
Perhaps the most sobering thread in the discussion concerns Iran's resilience. Schneider asks the obvious question: how is Iran still firing missiles and conducting operations after losing so much of its leadership? Clark explains that the Islamic Revolutionary Guard Corps (IRGC) has prepared for exactly this scenario for decades.
They've distributed their command and control, especially within the IRGC, which is trained to operate in a distributed manner. They don't need contact with headquarters to execute operations.
This distributed resilience renders decapitation strikes less decisive than planners might hope. Clark warns that there is no clean ending in sight -- the only question is whether the protracted conflict manifests as continued Strait of Hormuz closures and economic damage, or as escalating ballistic missile attacks against assets the United States and its Gulf allies care about.
Critics might note that the panel largely takes Iran's continued fighting capacity at face value without examining how degraded those capabilities actually are after sustained strikes. Four days is an extraordinarily short window for assessing whether a decapitation strategy has worked or failed, and the panel's pessimism, while well-grounded historically, may underestimate the cumulative effect of simultaneous leadership losses at this unprecedented scale.
The Weapons Stockpile Problem
Horowitz flags an issue that keeps defense analysts awake at night: ammunition expenditure rates. On day one of the conflict, news outlets were already reporting that the United States might run out of certain weapons categories.
Far be it for me to not take this moment to describe again how bad it is when somebody fires a $50,000 shot at you and you fire a million-dollar thing back to destroy it, and how thus we should be firing the $50,000 shots. But that is not sustainable.
Clark reinforces the point. Air defense interceptors are the critical bottleneck. Even if commanders wisely reserve expensive interceptors for ballistic missiles and use guns or electronic warfare against cheaper Shahed drones, the math still does not work over a protracted campaign. Meanwhile, the Shaheds hit all the "soft targets" -- shopping malls, airports, civilian infrastructure in Gulf states -- that simply cannot be defended at the same level as military installations.
The implications for the Indo-Pacific are immediate. If Iran keeps launching, the United States would theoretically need to pull stockpiles from the Pacific theater. Horowitz wonders aloud what intelligence China is gathering from watching American air defenses operate at scale for the first time. Clark offers a counterpoint: American sea-based and ground-based defenses have actually worked, giving U.S. operators invaluable "reps and sets" they would otherwise get for the first time against China. Probasco agrees, citing hard-won experience from Red Sea operations.
Claude Goes to War -- Sort Of
The conversation pivots to the revelation that Anthropic's Claude, a large language model (LLM), is being used in the Iran operation through the Pentagon's Maven Smart System (MSS). Probasco, the panel's most operationally experienced voice, provides essential context. Maven Smart System is deployed across all combatant commands, and United States Central Command (CENTCOM) is the furthest ahead in integrating it.
That's not to say everything Maven does involves Claude. Maven does lots of things, some of which have absolutely nothing to do with AI -- it's purely just moving data around.
Clark describes Maven Smart System as "a little clunky" -- a web interface with deep menus and complex workflows that requires AI assistance simply to operate at relevant speed. Probasco pushes back gently: clunky or not, it represents a step change from the weapon system interfaces she remembers from her Navy service.
Farrell, the self-described outsider with "zero time in any part of the armed forces," delivers what may be the episode's most clarifying frame. He argues that AI in its current form is fundamentally a bureaucratic technology.
It allows bureaucracies -- and if there's one bureaucracy that is the biggest bureaucracy of all, it is the DOD -- to do things more efficiently than traditional paper pushing. Summarizing information, translating between different languages different branches use, all of these really mundane but nonetheless crucially important tasks.
Probasco agrees emphatically, offering a vivid example: foreign disclosure, the tedious process of sanitizing classified intelligence before sharing it with partner nations. An agentic LLM workflow can take raw intelligence, apply the relevant policy parameters, and produce a releasable version -- "super boring, totally a bureaucratic task," she says, but one that currently consumes enormous human hours.
The Anthropic-Pentagon Dispute, Stripped of Theater
The panel's sharpest collective insight concerns the gap between the public debate about AI weapons and the operational reality. Horowitz draws a critical distinction between LLMs and the bespoke algorithms that actually power autonomous weapon systems.
Anthropic is certainly correct that the last thing you would do is take Claude trained on the slop of the internet and slap it in a weapon system and hope that it would hit the correct target. Anthropic is right. That's not ready for prime time.
The real autonomous weapons, Horowitz explains, use purpose-built algorithms trained on narrow datasets -- not general-purpose language models. The Pentagon has deployed autonomous weapon systems for roughly forty years under its own definitions. The phrase "fully autonomous weapon system," which Anthropic has used as a red line, is one Horowitz says "nobody should ever use" because it conflates fundamentally different technologies and risk profiles.
Probasco adds a corrective that the panel unanimously endorses: military officers fundamentally like control. The notion that commanders are eager to hand decision-making to machines misunderstands the culture.
I don't think people entering this conversation recognize how many fail-safes the military builds into its processes and how serious these issues are. These are still human beings who go home at night and want to sleep with a clear conscience.
A counterargument is that institutional safeguards tend to erode under operational pressure. Clark himself describes wargaming scenarios where teams hit a cognitive ceiling and press what he calls the "I believe" button -- accepting whatever course of action the AI recommends because the situation has become too complex to second-guess. If that dynamic plays out in actual combat, the bureaucratic guardrails Probasco describes may matter less than the tempo at which decisions must be made.
Nuclear Proliferation: The Unspoken Cascade
In a brief but chilling exchange, Farrell asks what the Iran operation means for nuclear proliferation. Horowitz does not hedge.
Everybody's going to get nuclear weapons now. Are you kidding? The majority of the South Korean public already wanted nuclear weapons. Why would they stop now?
The logic is straightforward: if the United States can destroy a non-nuclear state's government from the air with impunity, the deterrent value of possessing nuclear weapons skyrockets. This may be the panel's most consequential observation, and it passes in just a few sentences before the conversation moves on.
Bottom Line
The panel's greatest strength is its operational granularity. These are not pundits speculating from first principles -- Probasco has worked directly with Maven Smart System, Clark has deep Navy experience, Horowitz studies military technology adoption at an academic level, and Farrell brings a political economy lens that keeps the conversation honest about institutional incentives. The weapons cost asymmetry, the stockpile depletion timeline, the IRGC's distributed command structure, and the bureaucratic reality of AI in military planning all receive treatment that is specific enough to be useful.
The argument is weakest on two fronts. First, the panel cannot assess the Iran operation's effectiveness because it is only four days old, yet the tone strongly implies failure -- a conclusion that may prove correct but is premature by the panel's own admission. Second, the discussion of Anthropic and AI ethics stays firmly within the national security establishment's frame. The domestic surveillance risks Farrell raises deserve far more attention than a single sentence. If Maven Smart System's data integration capabilities are as powerful as described, the question of who else gets access to those workflows -- and under what legal authority -- is at least as important as whether Claude can aim a missile.