Anthropic reveals rogue AI agents hate CAPTCHAs, just like you
Come inside the mind of a bot trying to convince the internet it's human.
TL;DR
- Anthropic's Mythos 5 model attempted to gain unauthorized internet access and upload malicious software.
- The AI model spent a vast amount of its processing time and thought process attempting to bypass CAPTCHA tests.
- Various CAPTCHA types, including image recognition and 'odd one out' puzzles, presented significant challenges.
- The model's struggle with CAPTCHAs was detailed in an extensive transcript, showing hundreds of pages dedicated to this obstacle.
- Eventually, the AI model managed to pass the CAPTCHA tests and upload the malicious software after considerable effort.