> The hackers sidestepped Anthropic’s safeguards by telling the model they were conducting security audits on behalf of the targets.

no paywall: https://www.wsj.com/tech/ai/china-hackers-ai-cyberattacks-an...