“Sixty seconds is all it took,” said Oded Ailam, a former head of the Israeli intelligence apparatus Mossad’s counter-terrorism division and a researcher at the Jerusalem Center for Security and Foreign Affairs, in early March, describing the clinical Israeli-US operation that killed former Iranian supreme leader Ayatollah Ali Khamenei (1939 - 2026) on the first day of the joint military campaign launched by Washington and Tel Aviv against Iran on February 28.
The joint campaign, dubbed by the US as Operation Epic Fury and by Israel as Operation Roaring Lion, saw US Central Command forces strike more than 13,000 targets across Iran before it ceased under a two-week ceasefire mediated by Pakistan on April 8. For its part, Tel Aviv said it struck over 4,000 targets across Iran during roughly the same period.
The human toll of the six-week war reached staggering levels, with the Iranian government and the US-based Human Rights Activists News Agency (HRANA) reporting in early April that between 3,468 and 3,636 Iranians were killed during the conflict. Tehran’s health ministry also reported around 26,500 civilian injuries.
However, behind the casualty numbers lies a much bigger driver of the war, as it was commanded, in significant part, by two artificial intelligence (AI) systems - Maven and Mythos - which have seemingly rewritten the rules of modern warfare.
For the first time in history, the gap between identifying a target and striking it was compressed from hours to mere seconds, driven by the lethal pairing of these two AI systems. This has prompted military analysts to dub the US-Israel campaign against Iran the first AI war, a label underscored by the high casualty figures.
Maven and Mythos
Maven Smart System, developed by Palantir Technologies (PLTR) - a massive AI operating system powerhouse based in Florida - served as the nervous system and the platform that consolidated thousands of data streams into a single target factory. Maven ingests data from more than 150 sources, including drone video footage, geolocation data, infrared sensors, radar, satellite imagery, and signals intelligence, all fused in real time.
Highlighting the major turning point, Cameron Stanley, the Pentagon’s Chief Digital and AI Officer, remarked in mid-March, “We’ve gone from identifying the target to actioning that target - all from one system. This is revolutionary.”
A late-March report from The Week magazine further noted that Maven also carries an AI asset tasking recommender that proposes which bombers and munitions should strike which targets, and can apply automated legal reasoning to assess the grounds for a strike.
In the Iran war’s first 24 hours alone, Palantir Technologies’s Maven Smart System reportedly helped US forces identify 1,000 Iranian targets, according to an early-March report from The Washington Post. The report further noted that after integrating large language models, Maven’s processing rate increased to 5,000 targets per day, up from fewer than 100 before AI integration.
Palantir Chief Technology Officer Shyam Sankar remarked in mid-March that during the US-led invasion of Iraq in 2003, targeting a thousand sites required six months of work by fifty to a hundred analysts. However, in the 2026 Iran war, the equivalent was done by one person in two weeks.
Meanwhile, the Mythos AI model from Anthropic - a frontier AI research company best known for its Claude family of large language models (LLMs) - offered advanced reasoning capabilities and strong performance in cybersecurity-related tasks.
During the infamous 60-second strike that killed Iran’s former supreme leader on February 28, Mythos reportedly analyzed the pattern-of-life data and intercepted communications, predicting the exact moment Ali Khamenei would be above ground. Simultaneously, it identified and exploited thousands of vulnerabilities in Iranian communication networks, effectively blinding Tehran’s security teams just seconds ahead of the strike.
Within the Maven architecture, Mythos functions as the cognitive core, reading raw intelligence and reasoning about it. An analyst types a question in plain English; Mythos synthesizes satellite feeds, signals intercepts, geolocation data, and prior strike assessments, then returns a prioritized, confidence-weighted answer within seconds.
Mythos is, in short, positioned as the strategist that helps inform where to aim and why, with analytical depth that would take a human intelligence team days to produce, but does not itself execute strikes.
According to Wired, a major American publication covering emerging technologies including their impact on security, Anthropic’s Claude model operates within Palantir’s Impact Level 6 classified setup, the highest tier authorized for secret-level data.
Human cost
Amid the high human costs, the ethical stakes surrounding AI-assisted targeting in the Iran war became impossible to overlook. This was especially the case after a February 28 military strike on Shajareh Tayyebeh Primary School in the city of Minab, in southern Iran’s Hormozgan Province, killed at least 175 civilians, mostly children.
While the US initially denied involvement in the Minab bombardment, a leaked CENTCOM investigation reported by The New York Times in early March suggested the strike was a “likely mistake” caused by outdated targeting data within the Maven system, which failed to flag that a former military compound had been converted into a school.
Further highlighting the error, during Palantir’s Artificial Intelligence Platform Conference (AIPCon) in mid-March, a Pentagon official displayed a Maven map showing a target marker positioned over Minab.
Against the backdrop of the Minab strike, Paul Scharre of the Washington DC-based Center for a New American Security think tank in mid-March warned that “AI gets it wrong,” stressing, “We need humans to check the output of generative AI when the stakes are life and death.”
Palantir’s Maven in May evolved into a more permanent Pentagon program, with funding significantly increasing from $480 million to $1.3 billion, alongside broader Army digital modernization initiatives. Anthropic’s Mythos system, meanwhile, remains restricted, though its capabilities have been proven at lethal scale in live conflict zones.
The latter developments have fueled an ongoing debate about the future of warfare more broadly, particularly over how any potential renewal of hostilities in the region could arrive pre-loaded with targeting architectures and AI-driven kill chains already tested in Iran.
Accordingly, the question of who controls those systems - and what values endure under the pressure of live operations - is likely to remain a defining geopolitical issue of the next decade, alongside broader questions about how much autonomy should be delegated to AI systems in life-and-death decision-making, and what safeguards are required to maintain human accountability in active operations.