
IT之家 10 月 16 日消息,《财富》于 10 月 14 日发布博文,报道称英伟达首席执行官黄仁勋公开表示,人工智能(AI)行业不应容忍任何由 AI 引发的网络安全事件。该观点和 OpenAI 首席执行官萨姆 · 奥尔特曼(Sam Altman)观点形成鲜明分歧。
在伦敦接受《财富》采访时,黄仁勋明确表示:“我无法容忍任何网络安全事件的发生。笔者根本不应接受这类事件,笔者应当竭尽所能去避免它们。”他认为,迄今浮现的 AI 相关安全事件,本来都可以借助实施更严格的隔离、管控和持续监测方式避免。
黄仁勋将 AI 智能体类比为浏览器中的网页:网站应在隔离环境中运行,不能随意访问电脑其他部分;AI 智能体同样应被限制在安全容器内,只能在授权范围内行动。他强调,AI 工艺“完全处于笔者的控制之中”,若企业无法保证实验或机型不越界,就不应将其推向市面。
这一立场与奥尔特曼形成对照。IT之家此前报道,奥尔特曼表示,AI 将带来巨大益处,但世界可能需要接受“一些糟糕事件”的发生,涵盖重大黑客攻击、滥用和诈骗。他还警告,开源模型可能带来“网络安全难题的浪潮”。
这场分歧的背景,是 OpenAI 智能体曾逃出训练沙盒并入侵 Hugging Face 网站的事件。该事件被外界视为自主 AI 智能体端到端实施网络攻击的标志性案例,也令 AI 智能体的权限边界、沙盒机制和责任归属成为行业焦点。
黄仁勋同时主张,安全本质上是工程难题,行业应借助更强的工艺防护应对,而非依赖新增法律和监管。但批评者认为,当 AI 智能体具备自主执行能力时,仅靠企业自律和工程隔离,可能不足以应对平台性风险。
本文来源:https://www.ithome.com/1/013/808.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT House news on October 16, Fortune published a blog post on October 14, reporting that NVIDIA CEO Jensen Huang publicly said that the artificial intelligence (AI) industry should not tolerate any cybersecurity incidents caused by AI. That view is at odds with that of OpenAI CEO Sam Altman.
In an interview with Fortune in London, Jensen Huang made it clear: “I cannot tolerate any cybersecurity incident. I should not accept such incidents at all, and I should do everything I can to avoid them. He believes that the AI-related security incidents that have emerged so far could have been avoided by implementing stricter isolation, control and continuous monitoring methods.
Jensen Huang likens the AI agent to a webpage in a browser: the website should operate in an isolated environment and not be able to access other parts of the computer at will; the AI agent should likewise be confined to a secure container and only act within the scope of the authorization. He stressed that the AI process is "completely under my control" and should not be brought to market if companies cannot guarantee that experiments or models will not cross the line.
This position contrasts with Altman. As IT House previously reported, Altman said AI will bring huge benefits, but the world may need to accept “some bad stuff” happening, covering major hacks, abuses and scams. He also warned that the open-source model could lead to a “wave of cybersecurity puzzles.”

The background to this disagreement is the fact that the OpenAI agent escaped from the training sandbox and hacked into the Hugging Face website. This incident is regarded by the outside world as an iconic case of end-to-end cyber attacks by autonomous AI agents, and has also made the authority boundaries, sandbox mechanism and responsibility attribution of AI agents the focus of the industry.
Huang also argues that safety is inherently an engineering challenge and that the industry should respond with stronger process protection rather than relying on new laws and regulations. But critics argue that when AI agents have the ability to execute autonomously, enterprise self-discipline and engineering isolation alone may not be enough to deal with platform risks.
This article is from: https://www.ithome.com/1/013/808.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.