Palantir co-founder asks people 'not to judge' the AI bombing of a school in Iran that killed 150 children

By Zendoric · September 27, 2026.

Joe Lonsdale, co-founder of Palantir, called last Friday for a more lenient judgment of the February attack on a girls' school in Minab, Iran, in which more than 150 people died, most of them minors. He said so at an investor event hosted by Reuters, before the agency's global defense editor, Joe Brock. The attack, with two Tomahawk missiles, occurred on the first day of the war the Trump administration launched against Iran and, according to Pentagon officials cited by Bloomberg News, was due in part to an excessive reliance on Palantir targeting technology.

'It's very easy to judge things from the outside,' Lonsdale said, as reported by Mother Jones. 'It could be that someone made 99 good decisions and then one bad one, and that led to something horrible.' He went further: he suggested that Iran's leaders might have 'deliberately disguised' some of those things 'to make them look like targets,' 'the same way the Palestinians and others do,' with the aim of 'gaining global sympathy' at the cost of the United States striking the wrong place. It is a serious accusation that Lonsdale did not back with any evidence, and one that contrasts with a UN committee report released last week suggesting the attack could qualify as a war crime, something that for now is an assessment by that body and not a judicially established fact.

When asked how tech companies should be held accountable 'if things go wrong,' Lonsdale's answer was vague: 'Maybe there should be some law about the companies operating here and what they're allowed to do with them.' That regulatory lukewarmness coexists with a detail documented by Mother Jones that is hard to overlook: as managing partner of the venture capital firm 8VC, Lonsdale has, according to Forbes, some $6 billion invested in defense technology companies. Palantir, according to the outlet Sludge, is seeing a surge in federal revenue under the Trump administration, including an Army contract worth up to $10 billion; Anduril, another company backed by Lonsdale and also funded by Peter Thiel, signed an Army contract in March with a $20 billion ceiling. His think tank, the Cicero Institute, and his lobbying arm, Cicero Action, are moreover linked to hundreds of thousands of dollars in lobbying spending on behalf of companies in his own portfolio, again according to that same investigation.

None of this is new in Lonsdale's discourse: in a text published on his blog in 2025 he argued that it is possible to be a 'perfectly moral' person working in the arms industry, and that society suffers from a 'default presumption of evil' toward those who make weapons. Without people willing to do defense's 'dirty work,' he wrote, 'we'd be speaking German. Or Russian. Or Chinese.'

Here is the underlying problem, beyond this particular episode: whoever benefits most from the absence of strict accountability for military AI providers is also the one asking for public understanding when that AI fails at the cost of civilian lives. It is not a subtle or hidden conflict of interest; it is explicit, written into the contract figures and into the investment portfolio of the person speaking. And when the answer to 'should there be laws?' is 'maybe,' coming from someone with so much influence over the sector, the phrase weighs less than the status quo it protects.

This is the sharpest edge of the dual-use AI debate we have been following in other contexts: the same data fusion and pattern recognition capability that speeds up a medical diagnosis or a supply chain becomes an instrument of harm when applied to military targeting without an equally sophisticated layer of accountability. The real risk today is not a hypothetical superintelligence out of control: it is how already-deployed systems, built and marketed by the same companies that then help set the rules meant to police them, are used under the pressure of war. Our underlying thesis —that well-governed AI can, in the long run, reduce catastrophic human error in life-or-death decisions— does not change. But that promise only holds up with independent oversight, verifiable transparency and real export/deployment controls, precisely what this episode shows is still missing. Without that, what we have is not governed technological progress: it is an armed industry explaining why it does not need to be governed.