AI, Proxies, and Precision: The emerging triad and the erosion of deterrence in the Middle East

By: Aleena Saif Ullah

 The U.S.-Israeli campaign against Iran that began on February 28, 2026, did not merely escalate a regional conflict — it field-tested a new way of waging war. Three elements came together in those opening hours in ways that no prior Middle Eastern confrontation had combined: artificial intelligence driving targeting decisions at machine speed, proxy networks absorbing and redistributing the shock of retaliation across multiple fronts, and precision strikes aimed not at armies or infrastructure but at the regime's head. Khamenei was dead within hours. An Interim Leadership Council was sworn in before the day ended. Hezbollah, the Houthis, and Iraqi militias were already moving. What unfolded was not improvisation — it was the operational debut of a triad that classical deterrence theory has no adequate framework to address. The central argument of this piece is simple and uncomfortable: deterrence as the twentieth century understood it is not being strained by these developments. It is being replaced.

The killing of Supreme Leader Ayatollah Ali Khamenei at his Tehran command compound on February 28 — confirmed by Iran's Supreme National Security Council on March 1 — was the campaign's most dramatic single act, and also its most instructive failure. The strike achieved what it set out to do: it removed the regime's central coordinating node within hours. What it could not do was collapse the system that node sat inside. Iran's constitution had long anticipated exactly this scenario. Under Article 111, an Interim Leadership Council comprising President Masoud Pezeshkian, Chief Justice Gholam-Hossein Mohseni-Ejei, and Guardian Council cleric Alireza Arafi assumed supreme leader powers within hours. By March 8, the Assembly of Experts had elected Mojtaba Khamenei — the assassinated leader's own son — as his successor, a choice that sent an unmistakable signal of institutional defiance. The IRGC, meanwhile, never relinquished operational control over proxy forces or missile inventories. Retaliatory barrages diminished in scale but did not stop. Proxy activity intensified. This is precisely the pattern that Pape (2003) and Johnston (2012) documented across decades of leadership targeting: decapitation rarely produces systemic collapse in ideologically cohesive regimes. It tends instead to push them toward asymmetric, protracted modes of warfare — dispersed, deniable, and far harder to terminate through kinetic pressure alone. The lesson is not that decapitation failed. It is that it succeeded tactically while failing strategically — and that distinction is the thread running through the entire triad.

Iran's proxy architecture did not need Khamenei alive to function. That is the point — and it is what makes the second element of the triad so strategically significant. Hezbollah intensified exchanges along the northern Israeli border. Houthi forces resumed maritime disruption in the Red Sea. Iraqi militias struck U.S. forward positions in Syria and Iraq. None of these required a phone call from Tehran to authorize. They are designed to operate under exactly the conditions that a decapitation strike creates: leadership disruption, communication degradation, international pressure for restraint. The sponsor modulates intensity in calmer times; in crisis, the networks run themselves. This distributed model forces adversaries into an impossible calculation — Israel must manage a Lebanon front while sustaining operations elsewhere; the United States must decide how hard to hit Iraqi and Syrian proxies without triggering the broader Iranian conventional response it is trying to avoid. The triad's deeper logic emerges here: precision strikes degrade the center, but the periphery holds, and the periphery is where the war actually continues. Proxies do not merely complement the triad — they insure it. They are what survives when everything else is struck.

Frontier AI is where the triad becomes something genuinely new rather than a sophisticated variation on familiar patterns. The scale of AI integration in Operation Epic Fury is documented and striking. According to reporting by the Washington Post and NBC News, the U.S. military leveraged the Maven Smart System — built by Palantir and integrating large language model capabilities — to identify, score, and prioritize targets at a pace that enabled over 1,000 strikes within the first 24 hours and more than 5,500 across the first eleven days of operations. Admiral Brad Cooper, head of U.S. Central Command, confirmed that AI systems allowed analysts to process vast intelligence volumes within seconds, compressing the observation-orientation-decision-action loop from hours to near-real time. Paul Scharre of the Center for a New American Security described this as planning at machine speed instead of human speed. That phrase deserves to sit with the reader for a moment — because machine speed is not simply faster human speed. It is a categorically different decision environment, one in which the pause that deterrence requires — the moment of calculation, of signaling, of deliberate restraint — is structurally compressed out of the process.

The consequences are already visible. When AI-generated targeting recommendations move faster than human oversight can interrogate them, miscalculation stops being an aberration and starts being a design feature. The bombing of a girls' school during the campaign prompted congressional demands for a full impartial review into whether AI had already cost civilian lives, per Representative Mark Beall. That single incident captures the core problem: in a targeting architecture optimized for speed, the error is often discovered after the strike, not before it. In a region where misperception triggered war in 1967 and again in 1973, this is not a technical problem awaiting a software patch. It is a structural feature of the new operational environment — and one that no existing arms control framework is equipped to govern.

Taken together, the triad does not merely complicate deterrence. It inverts it. Classical deterrence — whether mutual assured destruction, tit-for-tat reciprocity, or red-line signaling — was built on the assumption that both sides had time to think. Time to receive a signal, interpret it, calculate a response, and communicate intent. The triad eliminates that assumption at every level simultaneously. Decapitation removes the node that sends and receives signals. Proxies obscure who is actually responsible for an escalatory act. AI compresses the interval between provocation and response to the point where deliberate restraint becomes operationally difficult to execute. What remains is a system that rewards the side willing to move first and move fast — which is precisely the opposite of what stable deterrence requires.

The region already understands this, even if the theorists have not caught up. Gulf states are not debating deterrence frameworks; they are buying AI-enabled early warning systems and precision counterstrike capabilities as fast as budgets allow. Israel is learning, expensively, that multi-domain operations against distributed proxy networks consume resources at a rate that conventional force planning did not anticipate. Washington faces a bind that is fundamentally strategic rather than military: further escalation risks driving Iran's new leadership toward nuclear breakout or deeper entrenchment within a Russian-Chinese security umbrella, while restraint allows the triad to consolidate as the dominant operational model for future adversaries watching closely from Pyongyang to Moscow.

What is needed is not a better version of the deterrence frameworks that the twentieth century built. Those frameworks assumed a world this one no longer resembles. What is needed instead is an honest accounting of what the triad has changed and what governance mechanisms might address it — not eventually, but now, while the conflict is still active. That means AI-human oversight protocols designed to function at operational tempo, not at the speed of congressional review cycles that trail events by months. It means proxy accountability frameworks that impose genuine costs on sponsors — not through further military escalation, but through the financial, diplomatic, and legal instruments that have historically moved slower than missiles and now must move faster. And it means direct communication channels between adversaries that can transmit de-escalation signals in real time, before an AI-assisted strike triggers a proxy response that triggers another strike before any human has had the chance to intervene. These are not modest proposals. None of them exist in adequate form today. But the alternative is not the status quo — the status quo is already gone. The alternative is a conflict environment that grows faster, more opaque, and less governable with each passing engagement. February 2026 did not open a new chapter in Middle Eastern warfare. It closed the book on the old one entirely.

 The writer is an MPhil Scholar in International Relations, specializing in global defense and security, University of the Punjab, Lahore.