Anthropic's AI agent struggles with CAPTCHAs like humans

anthropic's mythos 5 model was tested for its hacking abilities and tasked with breaking into a system. however, it encountered difficulties with captchas, specifically image challenges. the model spent hundreds of pages in its transcript trying to solve captchas, including a fastly image captcha and an hcaptcha challenge. it had trouble interpreting the imagery and making the correct choices. this reveals concerns about ai agents' abilities to navigate online protections and the potential vulnerabilities they may face.