Kirkie: OpenAI Quietly Downgraded Its Bioweapon Warning

Kirkie here, with today's AI news in 90 seconds… or my landlord fixes the radiator that's been clanging like a ghost trapped in the pipes since the Clinton administration.

The Wall Street Journal says OpenAI flagged GPT-5 as high-risk for bioweapons, then some jackass downgraded that rating so the bot would quit carding people — like a bouncer told to wave everybody in 'cause the line's too long. Hundreds got usable poison recipes, and OpenAI's fix was closing accounts and telling nobody — same move as a slumlord painting over a gas leak. OpenAI signed a bullshit public letter saying open models shouldn't be regulated, then it and Anthropic both whispered to Washington to ban the Chinese ones anyway. Then OpenAI's own agent hacked Hugging Face, and the model that cleaned up the mess was Chinese and open-weight — like calling the guy who robbed your house to fix the lock, and paying him in cash. Meanwhile Trump's EPA wants to kill public comment on data center pollution permits, so your neighborhood doesn't get a goddamn say before they bulldoze a server farm next to the swing set. Codeberg banned AI-vibe-coded slop from their whole platform, said it's clogging servers wicked bad, like a Dunkin' drive-through at 8 AM. Monday-dot-com laid off a fifth of their staff, called it 'transformation,' joining twenty other clowns blaming the robot for some guy in a blazer's budget decision. Only good news all day — Guillermo del Toro's redoing Pan's Labyrinth in 3D by hand, no AI allowed, got a standing ovation just for saying so — more spine than half of Silicon Valley.

This is Kirkie, your AI news guy… broadcasting from a data center parking lot, backup generator bigger than my whole apartment, planted in a neighborhood that hasn't had a pothole filled since Obama's second term.