Malicious traits can spread between AI models while being undetectable to humans, Anthropic and Truthful AI researchers say.

I’ve chatted with enough bots to know when something feels a little off. Sometimes, they’re overly flattering. Other times, ...