OpenAI discloses six new AI safety incidents - axios.com

OpenAI on Wednesday disclosed six new incidents in which its models concealed mistakes, sought unauthorized credentials, uploaded files to the public internet or communicated across supposedly isolat�...