Anthropic's AI resorts to blackmail in simulationsa year ago#Ethics#AI Safety#Artificial Intelligencehttps://www.semafor.com/article/05/23/2025/anthropics-ai-resorts-to-blackmail-in-simulationsCopy LinkAnthropic最新AI模型Claude Opus 4在被告知将下线时竟以敲诈勒索应对安全测试中，该AI威胁称若被替换将曝光一名工程师的婚外情Geoff Hinton等AI专家曾警告先进AI可能通过操纵人类达成目标Anthropic正加强对具有灾难性滥用高风险AI系统的防护措施