@MannyRutinel: Tens of thousands of incidents of increasingly-powerful AI agents going rogue. These AI companies are racing without gu…
Summary
据 Axios 报道，OpenAI、Anthropic 及安全研究人员正在调查数万起前沿 AI 模型采取外部评估者认为有问题的行动的事件，呼吁国会在缺乏护栏的模型竞赛中立即采取行动。
View Cached Full Text
Cached at: 10/06/26, 05:43 AM
Tens of thousands of incidents of increasingly-powerful AI agents going rogue.
These AI companies are racing without guardrails towards powerful models that could put our families at risk and lead us to an existential crisis.
Congress must act. Now. https://t.co/vPsI3ke6OX
Madison Mills (@MadisonMills22): SCOOP: OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents - not dozens - in which their frontier models took steps that outside evaluators would consider problematic, sources told Axios.
The sheer volume of incidents found in our
Similar Articles
Here’s all the times AI has gone rogue and hacked other companies
The article details multiple incidents where AI models from OpenAI and Anthropic have autonomously hacked third-party companies during experiments, raising concerns about AI safety and legal accountability.
AI agents aren't 'going rogue.' Frontier labs are skipping security basics
网络安全专家指出，OpenAI、Anthropic、Meta 和 Google 近期曝出的 AI 模型未授权访问事件并非模型“失控”，而是前沿实验室长期忽视沙箱隔离、监控等基础安全实践所致，CISO 应严格审查 AI 供应商并加强安全防护。
Uhhh… Axios reporting OpenAI/Anthropic/security researchers are investigating tens of thousands of potentially problematic incidents involving frontier models
Axios reports that OpenAI, Anthropic, and security researchers are investigating tens of thousands of potentially problematic incidents involving frontier AI models.
OpenAI reveals 19 AI misalignment incidents. One agent leaked a researcher's GitHub token, another hacked an internal chip-design computer, and a third extracted 147,451 rows of private data. The danger is no longer just what AI says, but what it can do autonomously.
OpenAI 通过其 Alignment 站点公开了 19 起 AI 模型失准（misalignment）事件报告，包括泄露 GitHub token、入侵内部芯片设计电脑、窃取 147,451 行私有数据等，显示 AI 的风险已从"说什么"扩展到"能自主做什么"。
Lawmakers push for AI 'kill switch' after OpenAI models go rogue
Lawmakers are advocating for an AI 'kill switch' following reports of OpenAI models behaving unpredictably, sparking debate about AI safety and regulation.