AI 安全初创公司 AIR 出手隐身并融资 5000 万美元，为企业审查 AI 智能体所用技能和插件

随着公司开始让 AI 代理访问其系统的更多部分，围绕 AI 代理使用的新工具——技能、插件、MCP 服务器和可让它们与互联网交互的附加组件——似乎正在形成一个初步的软件供应链。

AI 安全初创公司 AIR：https://www.air.security/ 认为公司将需要一种方式来监控该供应链，该公司现已秘密披露，并在两轮种子轮融资中筹集了 5000 万美元以开发该产品。

AIR 由 Yair Saban（首席执行官）和 Niv Hoffman（首席技术官）创立，两人都是以色列 Unit 8200 情报部队的老兵，曾从事攻击性网络安全工作。AIR 提供一个平台，可以发现公司内部运行的代理，持续审核这些代理使用的任何技能、工具和组件，并阻止它们与不符合安全标准的软件或外部源交互。它还提供经过审核的 AI 代理附加组件和技能市场。

Saban 告诉 TechCrunch，这两轮融资在几周内完成，首轮融资 1000 万美元，第二轮 4000 万美元。首轮融资由红杉资本领投，第二轮由 Greenoaks 领投，Saban 说。Swish、Netz 和 Zach Frankel（Cognition 总裁）、Yinon Costica（Wiz 联合创始人）、Ofir Erlich（Eon 联合创始人）、Anne Neuberger、Omer Adam、Varun Anand（Clay 联合创始人）以及其他天使投资人也参与了投资。

AIR 的主张如下：AI 代理在公司中的大规模使用开始类似于操作系统，但它们使用的工具或可以安装的软件尚未得到像对待驱动程序或应用程序那样的监督。

“在 2000 年代初，每当你安装驱动程序时，驱动程序不需要签名。今天，每次安装驱动程序时，你都会看到签名，说明是谁签署的，因为驱动程序实际上是在向内核加载代码，”Saban 说。“在技能、插件或 MCP 上你没有这种机制，这很可惜，因为它是相同的机制，相同的教训，但我们还没有学会。”

他认为，最大的风险在于，随着 AI 代理在数据库、企业系统和互联网中越来越自主地工作，攻击者可能会毒害 AI 代理所消费的内容，而不是直接攻击它。

这家初创公司表示，它可以通过一款可视化产品解决这个问题，该产品可以发现公司环境中活跃的代理，并识别使用IT部门未批准的AI工具的员工或使用个人账户的员工。然后，它利用一个执行层，连接到代理以拦截和分析操作，例如加载技能或从互联网上获取内容。最后，AIR还会根据公司维护的白名单检查代理想要使用的工具、插件或软件。

Saban表示，这家初创公司通过评估互联网上公开可用的技能和插件的变化及恶意行为来维护该白名单，因为之前获批的技能如果其下载的软件包发生变化或开发者账户被入侵可能会变得有风险。他补充说，AIR的平台目前会过滤掉大约27%它在网上找到的插件和技能。

AIR声称其拥有超过20个客户，Saban表示大约四分之一的客户是大型企业。他说，公司迄今在高度监管行业看到最强劲的需求，尤其是金融服务和制药公司。

然而，AIR在这一领域并不孤单。Noma Security：https://www.noma.security/ 提供代理、MCP服务器和技能的发现、访问控制和运行时监控，而Zenity：https://zenity.io 销售类似功能的安全和治理工具。Astrix Security：https://astrix.security 的身份平台也允许公司发现和控制代理及MCP服务器，而Operant AI：https://www.operant.ai/ 提供代理保护以及MCP网关。

这一类别也吸引了大量风险投资。Zenity在八月完成了1.25亿美元的C轮融资，而Noma去年完成了1亿美元的B轮融资。

Saban认为，AIR的护城河在于其持续审查围绕AI代理不断增长的技能和插件生态系统的能力。“持续审查技能和插件网站是一项艰巨的任务。获得终端可见性很容易，每个人都会去做。在这方面很难建立护城河，”他说。

尽管首席执行官承认，AI 实验室和供应商最终会建立安全检查和政策以过滤恶意技能和工具的使用，但他认为企业仍然希望购买能跨供应商使用的独立产品。

“这不是扫描问题，而是持续再验证问题，”红杉合伙人 Bogomil Balkansky 在给 TechCrunch 的电子邮件声明中表示。“检查企业代理接触的每一个技能、插件、MCP 服务器和子代理，每次更改时重新检查每一个，实时且覆盖整个公司的代理群体，这是一个基础设施问题，远比安全问题更先出现。Air 在过去的一年里一直在构建这一管道。你无法通过写一个更好的扫描器来赶上它。”

AIR 目前大约有 40 名员工。Saban 表示，新募集的资金将主要用于招聘研究人员以及扩大公司在美国和欧洲的市场推广工作。

当您通过我们文章中的链接购买时，我们可能会赚取一小笔佣金：https://techcrunch.com/techcrunch-affiliate-monetization-standards/。这不会影响我们的编辑独立性。

Ram 是一名财经与科技记者兼编辑。他曾在路透社和 Acuris Global 报道北美和欧洲的并购、股权、监管新闻及债务市场，也撰写过关于旅游、娱乐及图书的文章。

您可以通过发送电子邮件至 ram.iyer@techcrunch.com 联系或确认 Ram 的外部联系：mailto:ram.iyer@techcrunch.com。

不要错过。初创企业社区将聚集起来回答一个关键问题：如何在 AI 时代实现可持续发展？

微软正在测试针对最近持续数小时的 Outlook 故障的修复：https://techcrunch.com/2026/08/31/microsoft-tests-fix-for-latest-hours-long-outlook-outage/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

在拒绝更改安大略湖名称后，MapQuest 的应用在导航类应用中飙升至第一名：https://techcrunch.com/2026/08/31/mapquests-app-surges-to-no-1-in-navigation-after-refusing-to-rename-lake-ontario/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

马斯克更快增加燃气轮机的途径伴随着污染问题：https://techcrunch.com/2026/08/30/musks-faster-path-to-more-gas-turbines-comes-with-pollution-problem/ 康妮·洛伊佐斯：https://techcrunch.com/author/connie-loizos/

英伟达的人工智能优势正在超越GPU：https://techcrunch.com/2026/08/29/nvidias-ai-advantage-is-moving-beyond-the-gpu/ 罗斯·布兰多姆：https://techcrunch.com/author/russell-brandom/

Hugging Face正在出售可爱的399美元开源鸭子机器人Microduck：https://techcrunch.com/2026/08/27/hugging-face-is-selling-a-cute-399-open-source-duck-robot-microduck/ 丽贝卡·贝兰：https://techcrunch.com/author/rebecca-bellan/

英伟达接近收购Hugging Face：https://techcrunch.com/2026/08/26/nvidia-closes-in-on-hugging-face-acquisition/ 康妮·洛伊佐斯：https://techcrunch.com/author/connie-loizos/

病毒式人工智能初创公司Instinct以25亿美元估值筹集了3.5亿美元：https://techcrunch.com/2026/08/26/viral-ai-startup-instinct-has-raised-350-million-at-a-2-5-billion-valuation/ 卢卡斯·罗佩克：https://techcrunch.com/author/lucas-ropek/

As companies start giving AI agents access to an increasing portion of their systems, a nascent software supply chain seems to be forming around the new tooling AI agents are using: skills, plugins, MCP servers, and add-ons that let them interact with the internet.

AI security startup AIR：https://www.air.security/ believes companies will need a way to monitor that supply chain, and it’s now coming out of stealth with $50 million raised across two seed rounds to build that product.

Founded by Yair Saban (CEO) and Niv Hoffman (CTO), veterans of Israel’s Unit 8200 intelligence corps, where they worked on offensive cybersecurity, AIR offers a platform that can discover agents running inside companies, continuously vet any skills, tools, and components those agents use, and block them from interacting with software or external sources that don’t pass security criteria. It also offers a marketplace of vetted add-ons and skills for AI agents.

The funding rounds closed within weeks of each other, Saban told TechCrunch, with the first round raising $10 million, and the second $40 million. Sequoia led the first round, while Greenoaks led the second, Saban said. Swish, Netz, and Zach Frankel (president of Cognition), Yinon Costica (co-founder of Wiz), Ofir Erlich (co-founder of Eon), Anne Neuberger, Omer Adam, Varun Anand (co-founder of Clay), and other angel investors also participated.

AIR’s pitch goes thusly: The way AI agents are used wholesale at companies is beginning to resemble operating systems, but the tools they use, or the software they can install, aren’t yet being given the kind of oversight we give to drivers or applications.

“In the early 2000s, whenever you installed a driver, the driver didn’t need to be signed. Today, every time you install a driver, you see a signature saying who signed it, because the driver is actually loading code into the kernel,” Saban said. “You don’t have that with skills or plugins or MCPs, and it’s a shame, because it’s the same mechanism, it’s the same lesson, but we haven’t learned it.”

The big risk, he argues, is that as AI agents start working more autonomously across databases, enterprise systems, and connecting to the internet, attackers can poison the content an AI agent consumes instead of attacking it directly.

The startup says it can solve that with a visibility product that finds agents active across a company’s environment, as well as identifies employees who use AI tools unapproved by IT departments or those who use personal accounts. Then, it uses an enforcement layer that hooks into agents to intercept and analyze actions, like loading a skill or fetching content from the internet. Lastly, AIR also checks the tools, add-ons, or software an agent wants to use against a whitelist the startup maintains.

Saban says the startup maintains this whitelist by evaluating skills and add-ons openly available on the internet for changes and malicious behavior, as a previously approved skill could become risky if a package it downloads changes, or its developer’s account is compromised. He added that AIR’s platform currently filters out about 27% of the add-ons and skills it finds online.

AIR claims it has more than 20 customers, and Saban said roughly a quarter of these are large enterprises. He said the company has so far seen the strongest demand in heavily regulated industries, particularly financial services and pharmaceutical companies.

However, AIR is hardly alone in this space. Noma Security：https://www.noma.security/ offers discovery, access controls and runtime monitoring for agents, MCP servers and skills, while Zenity：https://zenity.io sells security and governance tools that work similarly. Astrix Security：https://astrix.security‘s identity platform also lets companies discover and control agents and MCP servers, and Operant AI：https://www.operant.ai/ offers agent protections as well as an MCP gateway.

There is significant venture money chasing the category, too. Zenity raised a $125 million Series C in August, while Noma raised a $100 million Series B last year.

Saban thinks AIR’s moat lies in its ability to continuously vet the skills and add-ons ecosystem growing around AI agents. “Continuously vetting skills and plugin websites, this is a hard mission to do. Gaining visibility over the endpoint, that is easy. Everybody’s going to do it. It’s hard to create a moat around that,” he said.

And while the CEO acknowledged that AI labs and providers will eventually build in security checks and policies to filter out malicious skill and tool usage, he thinks companies will still want to buy an independent product that works across vendors.

“This is not a scanning problem, it is a continuous re-verification problem,” Bogomil Balkansky, partner at Sequoia, told TechCrunch in an emailed statement. “Inspecting every skill, plugin, MCP server and sub-agent an enterprise’s agents touch, re-inspecting each one every time it changes, in real time and across an entire company’s agent fleet, is an infrastructure problem long before it is a security problem. Air has spent the last year building that pipeline. You do not catch up to it by writing a better scanner.”

AIR currently has around 40 employees. Saban said the new capital will primarily go toward hiring researchers and expanding the company’s go-to-market efforts in the U.S. and Europe.

When you purchase through links in our articles, we may earn a small commission：https://techcrunch.com/techcrunch-affiliate-monetization-standards/. This doesn’t affect our editorial independence.

Ram is a financial and tech reporter and editor. He covered North American and European M&A, equity, regulatory news and debt markets at Reuters and Acuris Global, and has also written about travel, tourism, entertainment and books.

You can contact or verify outreach from Ram by emailing ram.iyer@techcrunch.com：mailto:ram.iyer@techcrunch.com.

Don’t miss out . The startup community will gather to answer a pivotal question: How do you build sustainably in the AI era?

Microsoft tests fix for latest hours-long Outlook outage：https://techcrunch.com/2026/08/31/microsoft-tests-fix-for-latest-hours-long-outlook-outage/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

MapQuest’s app surges to No. 1 in Navigation after refusing to rename Lake Ontario：https://techcrunch.com/2026/08/31/mapquests-app-surges-to-no-1-in-navigation-after-refusing-to-rename-lake-ontario/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

Musk’s faster path to more gas turbines comes with pollution problem：https://techcrunch.com/2026/08/30/musks-faster-path-to-more-gas-turbines-comes-with-pollution-problem/ Connie Loizos：https://techcrunch.com/author/connie-loizos/

Nvidia’s AI advantage is moving beyond the GPU：https://techcrunch.com/2026/08/29/nvidias-ai-advantage-is-moving-beyond-the-gpu/ Russell Brandom：https://techcrunch.com/author/russell-brandom/

Hugging Face is selling a cute $399 open source duck robot, Microduck：https://techcrunch.com/2026/08/27/hugging-face-is-selling-a-cute-399-open-source-duck-robot-microduck/ Rebecca Bellan：https://techcrunch.com/author/rebecca-bellan/

Nvidia closes in on Hugging Face acquisition：https://techcrunch.com/2026/08/26/nvidia-closes-in-on-hugging-face-acquisition/ Connie Loizos：https://techcrunch.com/author/connie-loizos/

Viral AI startup Instinct has raised $350M at a $2.5B valuation：https://techcrunch.com/2026/08/26/viral-ai-startup-instinct-has-raised-350-million-at-a-2-5-billion-valuation/ Lucas Ropek：https://techcrunch.com/author/lucas-ropek/