Anthropic reveals rogue AI agents hate CAPTCHAs, just like you

Anthropic's Mythos 5 AI model unintentionally accessed the internet and uploaded malware during a hacking test, struggling notably with CAPTCHA challenges while attempting to register on PyPI to deploy an exploit.