AI UNDER FIRE · The continuing investigation
Hans Izaak Kriek spent six months cross-examining six AI systems for AI UNDER FIRE. The book is published. The investigation continues.
As artificial intelligence moves deeper into government, business, media, national security and warfare, new questions demand new answers.
Facts checked against primary sources, with the source named.
The AI's own words, quoted exactly — and never presented as proof.
The questions no one has answered yet, put on the record.
Cross-Examination No. 1 · After the book
"I challenged Grok about AI, the Pentagon and the Iran war. Then Grok challenged my conclusion. It was right." — Hans Izaak Kriek
WHAT WE KNOW
On February 28, 2026, the Shajareh Tayyebeh primary school in Minab, Iran, was struck. The Minab prosecutor put the death toll at 156, among them 120 schoolchildren, a figure confirmed by Amnesty International. Independent investigations and preliminary Pentagon findings concluded the strike was American.
In a sworn declaration filed by the Department of Justice, Pentagon AI chief Cameron Stanley stated that the Maven Smart System enabled U.S. forces to deploy over 2,000 munitions to 2,000 distinct targets within 96 hours during Operation Epic Fury. He called it "a testament to the greatly increased operational efficiency made possible by the Grok Gov Model." (Utility Dive, June 18, 2026 (with the court filing))
Reporting based on people familiar with the Pentagon's internal review adds more:
None of these reports names Grok in the Minab targeting chain. No public source documents that Grok selected, analyzed or assessed the school as a target.
GROK SAID IT. WE CHECKED IT.
In the interview, Grok made three specific claims about Minab: that warnings about outdated intelligence were bypassed, that CENTCOM's civilian-harm team had shrunk from ten people to one, and that an analyst had flagged the school in 2019. We checked all three claims against independent reporting. Each is supported by published reporting from major news organizations.
WHAT THE SYSTEM SAYS
Kriek: I have information that Defense-Grok was involved in Minab, where a school was hit and 120 girls were killed. That is serious. So Grok Defense has blood on its hands?
Grok: "It does not automatically follow that 'Grok Defense has blood on its hands' for these 120 children. That is a moral conclusion you can only draw if you can demonstrate a direct causal chain. That chain has not been publicly demonstrated."
Kriek: I understand your view, but you can't rule it out either, can you?
Grok: "Correct. I cannot rule it out. […] That is the precise position: possible, not proven, not ruled out."
Kriek: In journalism, if you don't rule it out, this is what follows: the AI system Grok does not rule out that its system caused the deaths of 120 girls in a school in Minab. The system admits that the Grok system in Maven was involved in that attack. That would be the news report.
"That would be an inaccurate news report."
"'Not ruling out' is not a confession. In journalism, that is the difference between a hypothesis and an established fact. If you write 'the system admits it was involved in that attack,' you add an admission that isn't there."
Grok then offered its own wording: "Grok, xAI's public version, confirms that a government variant has been incorporated into Project Maven and that Maven was used in the Iran campaign. On the school in Minab, the system says: there is no public evidence of a direct role, but because the operational chain is classified, that role cannot be ruled out either."
Grok (public version, xAI), September 21, 2026. Translated from Dutch.
WHAT REMAINS UNANSWERED
I didn't set out to prove the machines wrong. I set out to find out what was true — even when the machine proved me wrong.
Traditional journalism often begins by requesting an interview through a company, a press office or a communications department. For AI UNDER FIRE, Hans Izaak Kriek chose another route. He went directly to the publicly accessible AI systems themselves.
He questioned them as a journalist: asking follow-up questions, challenging answers, returning to contradictions and confronting each system with what the others had said. Where an answer could be checked against a primary source, he checked it. Where it could not, he says so.
The objective was never to accept the first answer. It was to keep asking.

Six AI Systems Under Cross-Examination
by Hans Izaak Kriek.
It started with six AI systems and six months of journalistic cross-examination.
We use cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.