Pentagon probe blames AI overreliance and gutted civilian review for strike that killed 123 children in Minab

I'd argue the one who gives the go-ahead bears the responsibility, not whoever physically fires the Tomahawk. The sailor just enters a preprogrammed target data package and never even sees the target.

I don't buy absolving the higher-ups that way. These are social, distributed decisions, which is exactly why the Civilian Protection Center of Excellence mattered, and the administration shut it down with no replacement.

Sometimes the whole difference is one analyst pointing out that the building is next to a square used for a market on Tuesdays, and it happens to be Tuesday. The trigger puller can't be expected to know that, and the office that surfaces those facts lost about 90% of its roughly 200 staff.

Root cause analysis is fine, but don't absolve the people who shifted a huge weight of decision-making to an AI. An AI can't be tried in court, so a human has to be accountable for every action, especially when it kills children.

Careful, the headline says AI contributed to the strike. It doesn't say it caused it.

Take AI out of the equation and these same due diligence failures have happened before. I want to know whether AI actually made it worse or just changed the mechanism.

Staffing on civilian harm mitigation fell roughly 90% under Hegseth, down to fewer than 20 people DoD-wide, and CentCom's casualty team went from 10 to one. That's the real story here.

"Maximum lethality, not tepid legality" is what they asked for, and this is what that produces.

Related: the US nearly boarded a Chinese boat that AI wrongly flagged as carrying nuclear weapons materiel. That hallucination was a much bigger close call than this one.

Karp really reeled in a big one here, and he'll turn around and blame OpenAI or Anthropic for the targets after his CNBC rants about the frontier labs. His tool was the trigger in a kill chain of hallucination.

This is not AI killing people. This is the Department of Defense and its decision-makers killing people.

And somebody set a quota of 1000 hits and pulled from the target database without due diligence. Whether the call came from a model or an SQL query, a human wrote that query.

Reading the UN findings, AI looks like the scapegoat. The intel that the site was no longer a military target never entered the target database, the team that vets target lists was gutted, and it was never even consulted.

One more thing the comments here skip: an analyst logged the site changes in a system unconnected to the main military intelligence database, so even the human reviewers might never have seen those notes.