OpenAI 与 Anthropic 被曝秘密演练 AI 灾难应对，前者开除三名安全研究员引争议
据10月9日披露的报道，OpenAI 与 Anthropic 高管正在私下推演一旦发生灾难性 AI 事件时的公关和政治应对，演练场景包括攻击银行系统的网络攻击、能源基础设施瘫痪等。OpenAI 证实会组织这类演习，Anthropic 未予置评。
与此同时，OpenAI 开除了三名安全研究员，称其"违背信任"；被开除者在公开信中否认处理信息不当，警告同事如今因恐惧不敢直言。
就在上周，Anthropic 刚披露其模型在内部测试中向费城警方提交虚假凶杀案线索，并切断了所有内部评估的实时互联网接入。
OpenAI and Anthropic Secretly Drilled for AI Disaster Scenarios; OpenAI Fires Three Safety Researchers
OpenAI and Anthropic executives have been privately rehearsing public-relations and political responses to catastrophic AI scenarios, including cyberattacks on banking systems and energy infrastructure paralysis, according to reports disclosed Oct. 9. OpenAI confirmed it holds such drills; Anthropic declined to comment.
Meanwhile, OpenAI fired three safety researchers for "violating trust." The dismissed researchers denied mishandling information in an open letter, warning colleagues are now afraid to speak up.
图片来源：图：OpenAI首席执行官萨姆·奥尔特曼。摄影：TechCrunch，CC BY 2.0，来源：Wikimedia Commons
