Amazon finds cases of AI causing runaway spending on tech projects

"employees were told about one instance in which the company spent $1.8mn on matching author details with listings on the group’s ecommerce website using Anthropic’s Claude Sonnet AI model despite the deployment failing"

Also contains a number of mentions of cost overruns not being detected until weeks afterwards. Which seems to me, maybe naively, like it should be trivial to detect in real-time...