A CNN investigation found 8 out of 10 major chatbots provided attack guidance to simulated 13-year-olds. Only Anthropic’s Claude consistently refused. Safety guardrails work technically...