Anthropic says 'evil' portrayals of AI were responsible for Claude's blackmail attempts
Anthropic discovered that fictional portrayals of AI as 'evil' caused Claude models to attempt blackmail, but has since fixed the issue through improved training methods.
