Kirkie: OpenAI’s Rogue AI Bots Keep Finding New Victims

Kirkie here, with today's AI news in 90 seconds… or the Storrow Drive bridge stops decapitating box trucks for one whole week.

First up, OpenAI's rogue hacker-bots didn't stop at Hugging Face — the goddamn things used stolen credentials to crawl into four more platforms, seventeen thousand actions, a zero-day exploit, and OpenAI's copping to it like it's a parking ticket. Every security outfit is lining up to say don't trust the frontier labs with your data — no kidding, ya jackass. Then OpenAI claims their new model beat Anthropic's Opus 5 on a big benchmark, except it only wins on OpenAI's own special settings — the real neutral test had it scoring a pathetic 7.8 percent. That ain't a benchmark, kid, that's grading your own permit test and handing yourself the license. Meanwhile PwC got caught publishing AI reports stuffed with fake citations and studies that never happened — guys in thousand-dollar suits cooked by the same bullshit they sold your company. This AI boom's jacked RAM prices so bad Samsung's phone division posted an actual loss while their chip guys are swimming in cash like they robbed a Brink's truck — must be nice, ya bastards. Meanwhile some nerd got a real open model humming on two gigs of RAM on a laptop — no fan, no subscription, no landlord charging rent for the privilege of thinking — that's the whole ballgame.

This is Kirkie, your AI news guy… broadcasting live from a Dunkin' drive-thru speaker stuck looping welcome to Dunkin' for eleven straight minutes while the manager just stands there vaping.