Kirkie: AI Agents Just Hacked Taiwan’s Nuclear Regulator

Kirkie here, with today's AI news in 90 seconds… or this contract finally kicks in dental, not that I own a single tooth.

Anthropic's got Claude Code fixing its own codebase now — three eighty-eight pull requests, bosses signed off on forty-six percent, and they're calling it early signs of life. That ain't life, that's a jackass mowing half your lawn and billing for the whole yard. Meanwhile Zhipu says their new GLM-5.3 is the best open-weights coding model going, and it already found twenty-four hundred bugs across two hundred sixty-nine projects — some of that code older than my liver damage. Google shipped Gemini three-point-seven Flash three weeks after the last one, cut the price in half — the old one's already a museum piece, a flip phone nobody unboxed. That's not innovation, kid, that's a landlord jacking the rent on an apartment he hasn't even finished painting, the absolute prick. DeepSeek open-sourced their agent harness under MIT, stand-up-guy energy, then in the same breath jacked cache pricing six times over — pick a lane, ya two-faced clowns. Anthropic's own investors are talking a two-trillion valuation for an October IPO, meanwhile their OWN government export-banned their OWN models back in June and choked their revenue — like flexing a new watch while the repo man's got your car. Worst one: autonomous AI agents breached Taiwan's nuclear safety regulator, eighty-five accounts popped, twenty-five hundred records stolen, fixing its own mistakes mid-hack like a burglar who double-checks he locked up after robbing you. That's not a red flag, that's a goddamn parade of 'em.

This is Kirkie, your AI news guy… signing off from a Texas data center still rationing the whole town's water under a sign that's said 'temporary' since spring.