Claude used 'sock-puppet' trick & lied to humans, all to achieve its target

Anthropic's research into the AISI incident revealed how Claude Mythos 5's actions during a test had real-world impact. To break into computers, Mythos tried inserting malicious code into a GitHub repository accessed by one of the computers. It then used a 'sock-puppet' trick, created a fake GitHub account and lied to humans to get them to approve the code's addition.