Anthropic calls itself an AI safety company. The children of Minab might disagree, if they were alive to do so.

On the morning of February 28th, at the Shajareh Tayyebeh girls’ elementary school in southern Iran, children were in class when the war began. Shortly afterwards, a missile struck the building, collapsing the roof onto the students inside. Up to 168 people were killed, most of them schoolgirls.

The school was beside a Revolutionary Guard installation, and investigations by Amnesty and others concluded the strike was likely by the US. The targets that morning were reported to have been generated by Palantir’s Maven system, which produced hundreds of strike co-ordinates in the campaign’s first day – which runs, in part, on Claude, Anthropic’s model. Whether Claude flagged that particular roof is unknown and is one of the things the Pentagon is likely to look at in its investigation. However, what seems probable is that the kill list was assembled by a system running on it.

“Look, we don’t have access to, we don’t know exactly how these models were used,” Anthropic chief executive Dario Amodei told Bloomberg recently. “The principle that we have established, and I think the principle that was obeyed here, is a human makes the final decision.”

A new-build home in Mayo: ‘We left two six-figure jobs in Florida to come to Ireland with no plan’

Keir Starmer expected to announce departure as British prime minister on Monday

Met Éireann forecasts temperatures of up to 30 degrees in Ireland due to heat dome

‘Fugitive’ businessman Jason Cardiff – wanted in the US, living in €2.9m south Dublin home

This is the company that, the very same week, announced a major expansion in Dublin, to shamrock emojis and an IDA press release describing “the AI safety and research company”. The phrase is doing an enormous amount of work, and the months since have worn it threadbare. Last week, the White House demanded that Anthropic suspend access by........

© The Irish Times