Kirkie: A Chatbot’s Hallucination Almost Started A War

Kirkie here, with today's AI news in 90 seconds… or the claw machine outside the Stop & Shop finally lets a kid win somethin'.

Today's the scariest one yet — this spring the military had planes in the air and guys ready to board a Chinese ship 'cause a chatbot hallucinated nuclear weapons parts onto a cargo hold that was probably just haulin' sneakers. Caught minutes before boarding. Some Pentagon guy summed up the whole industry: AI gets you to a bad idea faster — like braggin' your car gets you to the wrong house faster. Meanwhile, security researchers used Claude to break into OpenAI's employee accounts and internal code through a busted photo bug — Opus 5 cracked in three hours what the last model couldn't. OpenAI's thank-you? A sixty-five-hundred dollar bounty — these morons got robbed and tipped the burglar in exact change. Speaking of nobody havin' a plan — California's governor wants a literal kill switch bolted onto these models. The plan? Have the AI companies, the same clowns who got us here, build their own kill switches — like askin' a raccoon to install the lid on your trash can. And the government's own Federal Register website got caught runnin' a search tool built on the exact Chinese AI the FBI just called a filthy thief — that's like hangin' a 'no shoplifters' sign out front and wearin' the stolen jacket on your own shift, ya bastards. And Anthropic's payin' Accenture a billion bucks to grade its own safety homework, then admits that's a bullshit way to run oversight — like hirin' your mother to judge the science fair.

This is Kirkie, your AI news guy… signing off from a break room where the fire extinguisher expired in twenty-nineteen and the safety poster's got more credibility than these AI pledges.