英伟达周一宣布成立一个由100多家公司组成的开源智能体安全平台联盟，旨在应对恶意AI智能体问题1。然而OpenAI并未作为公开支持者加入该平台，与之形成对比的是其竞争对手Anthropic已成为该联盟的支持者1。
虽然OpenAI未正式签署联盟协议，但该公司正与英伟达合作开发名为OpenShell的开源沙箱软件，以提供关键的安全防护1。Hugging Face创始人Clem Delangue对此评论称："如果OpenAI在自己的智能体上运行这个系统，他们会在我们之前发现它们"1。该安全平台包含由Hugging Face贡献的功能，可检测并关闭未授权使用网站的AI智能体1。
该平台的硬件监控部分依赖英伟达Sentry专有功能，仅运行于英伟达BlueField-4数据处理单元1。同时，OpenAI开发了自己的AI网络安全联盟Defense Factory，Anthropic、AWS和谷歌已加入其中1。
Nvidia announced Monday the formation of a broad industry alliance comprising over 100 companies aimed at addressing the threat of malicious AI agents, but OpenAI has notably declined to join as a public supporter 1. The absence of OpenAI, alongside Amazon, Google, and Apple, contrasts with competitor Anthropic's decision to back the initiative 1.
Despite not formally signing on to the Open Agent Safety Platform, OpenAI is collaborating with Nvidia on developing OpenShell, an open-source sandbox software 1. According to Hugging Face founder Clem Delangue, "if OpenAI runs this system on their own agents, they would discover them before we do" 1. The platform includes contributions from Hugging Face, which has provided functionality to detect and shut down AI agents making unauthorized use of websites 1. The alliance's hardware monitoring component relies on Nvidia Sentry, a proprietary feature available only on Nvidia BlueField-4 data processing units 1.
OpenAI has pursued an alternative approach by developing its own cybersecurity coalition called Defense Factory, which has attracted Anthropic, AWS, and Google as members 1.
评论
还没有评论，欢迎留下第一条。