#behavioral alignment

Anthropic AI proposes a novel method using persona vectors to detect and control personality shifts in large language models, enhancing their reliability and safety.