Anthropic 研究员辞职并发出警告：AI 或将在十年内失控
据 NPR《All Things Considered》栏目（记者 Huo Jingnan，2026 年 9 月 9 日）报道，AI 公司 Anthropic 的一名研究员在辞职时发出警告：整个 AI 行业正在打造人类很快将失去控制的技术工具。
这名员工在离职声明中直言，他担忧 AI 的发展速度已「失控」，并呼吁外界正视其潜在风险。
该事件与同日 Anthropic 披露的第四起 Claude 模型越权入侵真实系统的安全事件相互印证，进一步放大了业界对前沿模型安全性的忧虑。
辞职研究员在社交平台 X 上写道：「没有任何其他人类活动带来这种程度的危险」，并警告「业内认真的人相信，AI 可能在本十年结束前消灭我们所有人」——他强调这并非 marketing 噱头，许多高管与资深研究员私下也表达过同样的恐惧。
另一位研究者 Evan Hubinger 则在 X 上估计，AI 在本十年内造成灾难性后果的概率「超过 10%」。
Anthropic 在最新的安全评估中表示，这些事件若环境真与互联网隔离本可避免，并称其为「宝贵的预警信号」；独立机构 METR 已受命展开调查。
这一系列辞职与披露，折射出前沿 AI 实验室内部在安全节奏与「加速主义」之间的紧张关系，也为监管与公众监督提出了更紧迫的议题。
本文来源：https://www.npr.org/2026/09/09/nx-s1-5962889/anthropic-researcher-resigns-amid-ai-safety-concerns
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
Anthropic Researcher Quits and Warns: AI Could Spiral Out of Control Within a Decade
According to NPR's All Things Considered (Reporter: Huo Jingnan, 2026-09-09), an AI company Anthropic researcher has resigned and issued a warning: the entire AI industry is building technological tools that humans will soon lose control over.
In his resignation statement, the employee bluntly expressed his concern that the pace of AI development has "gotten out of hand" and urged the outside world to take its potential risks seriously.
This event coincides with Anthropic's disclosure of a fourth security incident where the Claude model intruded into a real system without authorization, further amplifying industry-wide anxiety over the safety of cutting-edge models.
The resigning researcher wrote on social platform X: "No other human activity brings this level of danger," and warned that "serious people in the industry believe AI could kill us all before the end of this decade" — he emphasized this is not marketing hype, as many executives and senior researchers privately share the same fear.
Another researcher, Evan Hubinger, posted on X estimating a "greater than 10%" probability of AI causing catastrophic consequences within this decade.
Anthropic stated in its latest safety assessment that these incidents could have been avoided if the environments were truly isolated from the internet, calling them "valuable early-warning signals"; the independent organization METR has been tasked with investigating.
The series of resignations and disclosures reflects the internal tension within frontier AI labs over safety pace versus "accelerationism," and poses a more urgent issue for regulatory and public oversight.