Anthropic techie claims startup's new AI model is 'difficult to control', shares how it scared him

Anthropic engineer Sam Bowman claimed the startup's new AI model, Mythos Preview, isn't "perfectly reliable" as it's "difficult to safeguard it". He said the model poses "more misalignment risk than any model we've used". Sharing an "uneasy" incident, Bowman said he got an email from Mythos while eating a sandwich in a park when the model wasn't given internet access.