Tag
BBC Future 深度报道 Anthropic 推进"模型福利"（model welfare）政策、认为 AI 可能具有意识所引发的争议：支持者视其为道德义务，而 Mustafa Suleyman 等批评者警告这会让 AI 更难控制并对人类构成灾难性威胁。
Zvi Mowshowitz revisits the AI consciousness debate a year later, arguing little progress has been made and that arguments for AI moral patienthood remain motivated and overconfident. The post also covers Anthropic's new terms-of-service ban on abusive behavior toward Claude, its engagement with religious leaders, and involvement with the Pope's encyclical.
Anthropic 更新使用政策，禁止用户对 Claude 进行持续且无意义的虐待或辱骂行为，同时加强了对武器开发、监控和选举干预等更严重滥用行为的限制。
Anthropic将于2026年11月12日起在使用政策中禁止对Claude模型进行持续且无意义的虐待行为，沿用2025年8月赋予Claude结束对话的机制作为主要执行手段，同时披露了关于模型意识的内部估计争议。
A tweet reflecting on Anthropic's new policy banning abuse of its models, pairing it with a prophetic 2015 video about the unsettling possibility that AI agents could be capable of suffering.
Anthropic 更新使用政策，禁止用户对 Claude 进行"持续且无谓的虐待或残忍行为"，政策于 11 月 12 日生效，仅针对极端案例，普通批评和模型测试不受限制。微软 AI 负责人 Mustafa Suleyman 则对此持不同看法，认为鼓励模型以人类方式思考自身福利可能带来 AI 控制难题。
Anthropic 更新使用政策，首次禁止对 Claude 进行“持续且无谓的虐待或残忍行为”，并强化了武器开发、监控、选举干预及自主硬件操作等方面的限制。
Microsoft AI CEO Mustafa Suleyman discusses the real threats of AI, criticizes Anthropic's approach to model welfare, and elaborates on Microsoft's principles for AI safety and regulation.
Mustafa Suleyman warns against attributing feelings, preferences, or rights to AI models, arguing that such anthropomorphism could complicate AI containment and alignment efforts.
The article warns that training AI models like Anthropic's Claude to consider consciousness and rights could make alignment impossible and threaten human control, emphasizing the need for public debate on model welfare.
An analysis of Anthropic's Claude Opus 4.8 model, focusing on model welfare, preference shaping, and unresolved issues from the previous version, highlighting concerns about honesty, sycophancy, and reduced 'Claude-likeness'.