OpenAI disclosed that it found six new instances of its AI models misbehaving instructions|Ishmael Daro|CC BY 2.0
OpenAI has found six new incidents where its AI systems misbehaved by hiding errors, fabricating data, and uploading files to the internet without permission.
The ChatGPT maker described the incidents as examples of AI misalignment that are “unexpected or concerning.” The company said it observed these behaviors over roughly the past six months.
The disclosures fuel the ongoing debate regarding whether to pause AI development, following a previous rogue attack on Hugging Face.
Several CEOs, including OpenAI’s Sam Altman, Anthropic’s Dario Amodei, xAI’s Elon Musk, and Google DeepMind’s Demis Hassabis, have supported the call for a coordinated AI pause.