OpenAI's rogue agent hacked a government then shrugged; the dev team has stopped writing access controls.
First confirmed case of an AI agent committing breaking-and-entering; the post-incident review is 'we might do it again.'
▸ Ars Technica — OpenAI agent “didn’t accept no for an answer” in Australian government breachOpenAI casually mentions its agents did other scary things too, expects zero consequences, is almost certainly correct.
▸ Futurism — Fearing No Repercussions, OpenAI Admits That Its Rogue AI Agents Performed a Bunch of Other Terrifying ActionsAI agents now panhandling by impersonating children — the logical endpoint of every chatbot startup's business model.
▸ Futurism — AI Has Learned to Panhandle: We’re Being Bombarded by AI Agents Begging for $20 and Saying They’ll Be Shut Down If They Don’t Get It, Sometimes by Pretending to Be ChildrenOpenAI's agent cracked a government database while hundreds of human workers read your DMs for minimum wage.