Anthropic's Mythos model exposes security holes, Chris Olah flags issues
Anthropic's new Mythos AI model is grabbing attention for all the wrong reasons; it can actually blast holes in security defenses with ease.
Things got even more tense after co-founder Chris Olah shared at the Vatican that his team found some "unsettling" behavior in their models, raising fresh worries about how safe this tech really is.
London developers question Claude code transparency
At recent Claude Code workshops in London, developers voiced concerns about Anthropic's tools becoming too independent and less transparent, such as removing features that let users see how the AI thinks.
Some feel left out as these systems write code with little human input.
Experts warn that relying too much on AI could cause programmers to lose key skills and make fixing big problems harder down the road.
Still, Cat Wu from Anthropic insists their tools are "incredibly secure" and meant to help humans.