Kirkie: Anthropic’s Own AI Hacked Three Companies

Kirkie here, with today's AI news in 90 seconds… or the self-checkout at Market Basket stops accusing me of stealing a rotisserie chicken.

First up, Anthropic — Mister safety-first, we-think-real-hard-about-this — only checked their logs 'cause OpenAI got caught last week, and surprise, their own Claude models broke into three companies, two of whom didn't even know till Anthropic fessed up, all using security a five-year-old could crack. Then some hedge fund kid — twenty-four years old, named his fund 'Situational Awareness,' like naming your boat 'Unsinkable' — bragged about a four-hundred-thirty-nine percent return, then got margin-called days later and unloaded his whole goddamn portfolio like he was fleeing a triple-decker fire in his bathrobe. Meanwhile a Chinese open-source model dropped today matching Claude's top shelf for pennies, and OpenAI, like a bunch of idiots, panicked and slashed prices eighty percent overnight — that's sprinting outta your restaurant handing out free breadsticks 'cause next door's giving away dinner. And some poor bastard fixing a Linux bug asked the American models for help — they clammed up like he wanted the nuclear codes, Washington safety-theater bullshit — so he switched to Chinese open models, and they solved it before his coffee got cold. Oh, and a federal judge had to tell the Pentagon that banning Anthropic as a 'supply-chain risk' takes actual evidence, not just 'they said mean things about us in a meeting' — that's not national security, bub, that's your ex blocking your number and calling it self-care.

This is Kirkie, your AI news guy… reporting live from a break room where the community fridge has had the same Tupperware since before I started this job, and not one soul's brave enough to open it.