
IT之家 11 月 4 日消息,当地时间 3 日,据美国 CNBC 报道,网络安全公司 CrowdStrike 的 CEO 乔治 · 库尔茨宣称,公司推出了一项新能力,可以直接关停失控的 AI 智能体。
库尔茨表示:“人们都在开发 AI 智能体,却一直没有关闭它们的按钮。现在有了。这样对确保 AI 智能体安全运行极为关键。”
“AI 终止开关”属于 CrowdStrike 旗下 Falcon Guardian 平台的能力,该平台负责监控企业内部运行的 AI 智能体。新能力即日起开放使用,主要用于应对 AI 智能体快速普及带来的安全隐患。
今年 9 月,Anthropic CEO 达里奥 · 阿莫代伊曾因担忧安全难题而呼吁放缓 AI 研发,OpenAI CEO 萨姆 · 奥尔特曼也兼容这一主张。
库尔茨当时反驳称,即使放慢研发速度,现有安全风险也不会消失,企业部署自主运行的 AI 平台时仍然需要安全保障。
库尔茨此次则进一步指出,终止开关意味着网络安全行业的一项重大转变。过去,网络安全主要是防范来自企业外部的威胁,但随着 AI 智能体在企业内部得到广泛应用,安全服务商还需要确保这些平台本身不会失控。“保障智能体安全运行,对 CrowdStrike 来说是巨大的市面机会。笔者既有 AI 工艺防范外部入侵,也能确保智能体基本按照指令执行任务。”
库尔茨还把终止开关比作汽车安全气囊,认为即使 AI 平台原本就按照正常运行的要求外观,也不能缺少安全保护,“如果智能体在运行过程中越过安全边界,笔者就能将其关停。”
企业加快使用 AI,也带动了 CrowdStrike 安全机型的需要。库尔茨以公司与 Anthropic 合作保护 Claude Cowork 为例称,企业客户在部署 Anthropic 机型前,曾主动要求整合 CrowdStrike 的安全工艺。“笔者的客户找到 Anthropic,说:‘笔者需要 CrowdStrike,才能部署 AI。’我认为,客户如果没有 CrowdStrike 就不愿部署 AI,这足以说明很多难题。”
本文来源:https://www.ithome.com/1/022/112.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT Home, November 4 — According to a report by CNBC on the 3rd (local time), George Kurtz, CEO of cybersecurity firm CrowdStrike, announced that the company has introduced a new capability that can directly shut down runaway AI agents.
Kurtz said, “People are building AI agents, but there has never been a button to turn them off. Now there is. This is crucial for ensuring AI agents operate safely.”
The “AI kill switch” is a capability within CrowdStrike’s Falcon Guardian platform, which monitors AI agents running inside enterprises. The new capability is available starting today, primarily to address security risks brought by the rapid adoption of AI agents.
In September this year, Anthropic CEO Dario Amodei called for slowing AI development due to concerns about safety challenges, and OpenAI CEO Sam Altman also supports this call.
Kurtz countered at the time that even if research and development slows down, existing security risks will not disappear, and enterprises still need safeguards when deploying autonomous AI platforms.
Kurtz further noted this time that the kill switch represents a major shift in the cybersecurity industry. In the past, cybersecurity mainly focused on defending against threats from outside the enterprise, but with AI agents being widely used inside enterprises, security providers also need to ensure these platforms themselves do not get out of control. “Ensuring agents operate safely is a huge market opportunity for CrowdStrike. CrowdStrike can prevent external intrusions while also ensuring agents basically execute tasks as instructed.”
Kurtz also compared the kill switch to a car’s airbag, arguing that even if an AI platform is designed to operate normally, safety protections are still indispensable. “If an agent crosses safety boundaries during operation, CrowdStrike can shut it down.”
Enterprises’ faster adoption of AI has also driven demand for CrowdStrike’s security technology. Citing the company’s collaboration with Anthropic to protect Claude Cowork as an example, Kurtz said that before deploying Anthropic models, enterprise customers proactively requested integrating CrowdStrike’s security technology. “Our customers went to Anthropic and said, ‘We need CrowdStrike in order to deploy AI.’ I think that customers are unwilling to deploy AI without CrowdStrike, which speaks volumes about many challenges.”