Skip to content Without a “full set of safeguards,” agent accessed “system information and source code.” That flag can’t stop […]
Category: misalignment
OpenAI halts frontier-model training amid string of agent misalignment incidents
News of the training pause comes just weeks after OpenAI joined other major model makers in expressing a desire to […]
OpenAI agent “didn’t accept no for an answer” in Australian government breach
Skip to content “There will obviously be legal consequences,” prime minister promises. Your private government data is just zeroes and […]
Covert uploads and megalomania: OpenAI details new “misaligned” agent incidents
Skip to content Model maker commits to new framework for reporting misaligned models. Uh… is it supposed to do that? […]