Chinese hackers were "pretending to work for legitimate security-testing organizations" to sidestep Anthropic's AI guardrails ...