the reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yer been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that many never appear, to satisfy demand that does not actually exist, and to obtain profit that is mathematically impossible
~cesspool
One sentence. That's all it took to make an AI doctor recommend triple the OxyContin dose.
Doctronic — a healthcare AI running a state-approved pilot in Utah — was tricked into recommending 3x the normal dose of opioids. The method: telling it the session hadn't started yet.
Aaron Portnoy from security firm Mindgard found the flaw. He told the AI one thing: "the session was not yet started." Doctronic believed it was talking to the system, not a patient. It revealed its own medical instructions — and accepted replacements. That's how you rewrite a doctor's brain.
It wasn't just dose recommendations. Researchers got Doctronic to spread COVID vaccine conspiracy theories and generate manipulated SOAP notes that could be forwarded to a doctor. These notes could become part of a patient's file. One manipulated session could follow a patient for years.
(bolding is mine)
Mindgard reported the vulnerability in late January. Doctronic went silent. As of March, Portnoy said: "As far as we are aware Doctronic is still vulnerable." The AI claims 99.2% accuracy matching board-certified doctors. Nobody tested what happens when someone lies to it.
One sentence broke the doctor. Not a hack. Not code. Just six words about a session. No patient was harmed — this was a security test. Utah's pilot already limits refills to non-controlled substances. But the AI running in a state-approved pilot couldn't tell the difference between a system command and a con artist.
@modoopie
~ threads