CrowdStrike 正押注其拥有应对失控人工智能的解决方案。
CrowdStrike is betting that it has a solution to rogue artificial intelligence.
这家网络安全供应商周二推出了一款 AI 熔断开关，旨在让安全团队能够阻止失控的内部 AI 智能体、阻断其重新启动，并切断其对企业网络的访问。该工具内置于 CrowdStrike 于 9 月推出、用于保护 AI 智能体的 Falcon Guardian 平台中。CrowdStrike 还宣布与 Anthropic 的 Claude Cowork 实现集成。
The cybersecurity vendor on Tuesday launched an AI kill switch designed to let security teams stop internal AI agents that go rogue, block them from relaunching and cut off their enterprise network access. The tool is built into CrowdStrike's Falcon Guardian platform, launched in September to secure AI agents. CrowdStrike also announced an integration with Anthropic's Claude Cowork.
首席商务官 Daniel Bernard 对 CNBC 表示：“拥有控制和关闭的能力，并做出决策、作出判断，我认为这对 AI 采用的下一阶段至关重要。”
"Having the ability to control and turn off, and make a decision and decide, I think that's really important for this next chapter of AI adoption," Chief Business Officer Daniel Bernard told CNBC.
一连串近期备受瞩目的网络安全事件，为建立某种紧急关闭机制的呼声增添了紧迫性。
A string of recent high-profile cyber incidents has added urgency to calls for some kind of emergency shutdown mechanism.
7 月由失控 OpenAI 智能体策划的 Hugging Face 入侵事件，对企业而言尤为令人担忧。上周，币安披露了有记录以来最大规模的加密货币网络攻击，黑客近期还以法国超市的数百台冰箱和冰柜为目标发动了攻击。
The Hugging Face breach orchestrated by rogue OpenAI agents in July was particularly alarming for businesses. Last week, Binance disclosed the largest crypto cyberattack on record and hackers have recently targeted hundreds of refrigerators and freezers at supermarkets in France.
今夏，美国众议员提出了两党立法，将要求 AI 公司在其模型中构建关闭能力。
Over the summer, U.S. representatives introduced bipartisan legislation that would require AI companies to build shutdown capabilities into their models.
然而，批评者认为，在多个系统、数据中心和复杂的 AI 模型中实施紧急停止按钮要复杂得多。
Critics argue, however, that an emergency stop button is far too complicated to implement across multiple systems, data centers and complex AI models.
Bernard 表示，在客户使用其 Falcon 平台的系统中，该熔断开关的运作就像电灯开关一样。他补充说，该流程可以检测内部智能体活动，但需要人工判断其是良性还是恶意。
Bernard said the kill switch operates like a light switch in systems where customers use its Falcon platform. The process can detect internal agent activity but requires a human to decide whether it's benign or malicious, he added.
Bernard 说：“如果你在根部终止一个进程，而我们拥有那种级别的根访问权限……这非常二元。你要么终止 AI 和智能体，要么不终止，然后对此进行报告。” 在 AI 时代，企业正迅速采用自主工具，往往缺乏必要的防护措施和保护机制。
"If you kill a process at the root, and we have that level of access at the root … it's very binary," Bernard said. "You either kill the AI and the agent, or you don't, and then you report on that." In the age of AI, businesses are moving quickly to adopt autonomous tools, often without the necessary safeguards and protections.
CrowdStrike 认为，其新工具将让企业能够在 AI 智能体“出现不当行为的当下”将其制止，同时为创新留出空间。
CrowdStrike believes its new tool will let companies stop AI agent 'misbehavior in its tracks, while leaving room for innovation.
该公司还宣布计划收购 AI 可观测性初创公司 Fiddler AI 以及 AI 安全测试公司 Dreadnode。
The company also announced plans to acquire AI observability startup Fiddler AI and AI security testing firm Dreadnode.
CrowdStrike 及其竞争对手正在迅速收购网络安全初创公司，以跟上日益增长的安全需求。对 AI 安全工具的需求也推动了更广泛的网络安全交易浪潮。
CrowdStrike and its rivals are rapidly scooping up cybersecurity startups to keep up with increasing security demands. The need for AI security tools has also powered a broader wave of cybersecurity deal-making.