AI News

前 Unit 8200 成员创办的 AI 安全公司 Bay 九个月融资 5000 万美元

CTech更新于 11月16日 08:47

Former Unit 8200 officers raise $50 million in just nine months for AI security startup

AI 导读

以色列 AI 安全初创公司 Bay 结束隐身状态,成立九个月内累计融资 5000 万美元,其中 3700 万美元 A 轮由 Battery Ventures 领投,Micron Ventures、Microsoft 旗下 M12 和 Wiz CEO Assaf Rappaport 参投,此前还有 1300 万美元种子轮。公司由 Yuval Zacharia、Aner Morag 和 Boris Danilovich 于 2026 年初创立,产品用于识别企业内部运行的 AI 智能体,并结合其任务和会话历史在毫秒级判断放行、阻断或转人工审批。Bay 称已服务 Wix、ClickUp、Micron、Peloton 等数十家企业客户,平台可在五分钟内基于现有管理设施部署,支持 AI 助手、编码智能体和 MCP 服务器。

推荐理由

AI 评分 57

这家公司针对 AI 智能体越权操作的风险做实时拦截,创始团队背景和早期客户数据可作为观察智能体安全赛道的样本。

主题

标签

正文

Bay, a cybersecurity company developing a platform to prevent potentially dangerous actions by AI agents within organizations, has emerged from stealth with $50 million in funding raised within nine months of its founding. The financing includes a $37 million Series A led by Battery Ventures, with participation from Micron Ventures, M12, Microsoft's venture fund, and Assaf Rappaport, CEO of Wiz. Micron is also a customer of the company. The round follows a previously undisclosed $13 million Seed investment led by YL Ventures and Striker. The company's private investors also include Jeetu Patel, CPO of Cisco, and Manoj Apte, former chief strategy officer at Zscaler.Full list of Israeli high-tech funding rounds in 2026Bay was founded in early 2026 by Yuval Zacharia, its CEO; Aner Morag, its chief product officer; and Boris Danilovich, its VP of R&D. The founders say the idea for the company emerged after one of their coding agents attempted to expose a cloud environment to the internet while trying to complete a test. The agent had not been compromised and was using legitimate permissions to pursue an assigned task, but its actions could have created a serious security risk.

1 View gallery Bay founders. (Photo: Arnon Harush)Zacharia began her career in Israeli intelligence, serving in Unit 8200, where she led a department of approximately 50 people. She subsequently became the first intelligence officer to transition into offensive cybersecurity and was assigned to one of the unit's most elite departments. After her military service, she served as director of research and AI at Hunters, where she led the development of the company's AI capabilities. Morag moved from intelligence roles into offensive cyber operations within Unit 8200 and joined Noname Security among its earliest employees, eventually rising to VP of Product. Danilovich was among the first employees at Preempt Security, which was acquired by CrowdStrike. He later led a department of approximately 90 engineers at CrowdStrike, working on a key component of its endpoint security platform and serving as a senior leader at the company's Israeli R&D center.Bay employs approximately 35 people, most of them in Israel. The company plans to expand its product and engineering teams, while growing its sales, marketing, and customer teams in the United States and Europe. Zacharia also intends to relocate to the United States."We will continue to build the product, expand into new areas, and help organizations adopt it rapidly, enabling them to deploy it comprehensively across the enterprise. We plan to grow our sales team, and I intend to relocate to the United States," Zacharia says.In a conversation with Calcalist, Zacharia said the company was able to raise capital quickly because it had already attracted a substantial number of customers and generated significant revenue in a short period."We raised a significant amount of capital, and we did so very quickly. The primary driver was the large number of customers we managed to acquire and the revenue we generated in such a short timeframe, which helped us close the funding round rapidly. Our approach to building the product is technologically distinct. From the moment we entered the market, we generated millions in revenue within the very first quarter of sales."Zacharia said the company had prioritized ease of deployment and the ability to demonstrate value to customers within hours. It also sought to make the case for rolling out the platform across an entire organization, rather than limiting it to individual teams or applications."Our product is designed to facilitate work with autonomous agents and monitor them wherever they operate; it is not merely a point solution. If an agent is about to take an action that constitutes a violation, the system intervenes to stop it," she said."There are a great many agents performing actions that are either malicious or questionable. Often, this isn't because a 'villainous' hacker has launched an attack; rather, many of these agents are simply striving to fulfill their assigned tasks, but in the process, they make mistakes, creating security issues that aren't always easy to resolve," Zacharia said. "During my military service, there was a strong emphasis on 'capability,' the idea that anything is possible. Similarly, AI agents will do whatever it takes to successfully complete a task, even if the cost to the organization is significant."Bay says it serves dozens of enterprise customers worldwide in sectors including finance, insurance, technology, manufacturing, and media. Its customers include Wix, ClickUp, Micron, and Peloton. The company says some of its Fortune 100 customers have tripled their use of autonomous AI agents while maintaining their security standards.

Related articles:

Israeli cyber startup Daylight raises $80 million Series B at estimated $400 million valuation

Spines raises $64 million to automate book publishing with AI

Ex-Unit 8200 commander's Sentra raises $150 million to secure corporate data as AI agents proliferateThe platform identifies AI agents operating within an organization and evaluates their proposed actions in the context of the full agent session, taking into account both the task they have been assigned and their previous activity. This allows it to distinguish between actions that may appear identical in isolation but carry different levels of risk depending on the agent's purpose and what it has done before. Within milliseconds, the system decides locally whether to allow an action, block it, or send it for human approval, according to the organization's security policies. Approved actions proceed without interrupting the agent's work.Bay says its system can be deployed in less than five minutes using an organization's existing management and security infrastructure, without requiring an additional persistent endpoint agent, kernel driver, or network proxy. It supports AI assistants, coding agents, and custom-built agents, as well as the MCP servers and tools they use. The platform combines AI discovery and security posture management, session-level monitoring, and real-time enforcement under a unified set of policies across employee devices, cloud infrastructure, and SaaS applications. It also connects to existing security workflows to support automated remediation. The company says its technology adapts to an organization's agent activity, reducing the need for security teams to create and maintain separate rules for every new agent or tool.

来源

CTech