Anthropic holds back 'Claude Mythos Preview' over cybersecurity breach risk
Technology
Anthropic, the AI company, is holding back its latest model, Claude Mythos Preview, because it's just too good at finding software flaws.
former CEO Matt Shumer told CBS Mornings the tech could break into almost any software, even things like power grids and banks, raising major cybersecurity concerns.
'Claude Mythos' access limited to operators
Since Claude Mythos is much better at hacking than earlier models, Anthropic is only letting entities responsible for securing critical infrastructure use it.
This gives institutions time to strengthen defenses before similar AI tools show up elsewhere in the next year.
Shumer also mentioned worries about other countries racing ahead with this kind of technology.