Bay has emerged from stealth with $50 million in total funding to address one of the biggest security challenges created by the rapid adoption of autonomous artificial intelligence: controlling what AI agents can actually do inside enterprise systems. Announced on November 10, 2026, the financing comprises a previously undisclosed $13 million seed round led by YL Ventures and Striker Venture Partners, followed by a $37 million Series A led by Battery Ventures. Micron Ventures, Microsoft’s M12 venture fund and Wiz CEO Assaf Rappaport also participated in the Series A. Based in Tel Aviv, Bay was founded in 2026 by Yuval Zacharia, Aner Morag and Boris Danilovich. Its platform is designed to discover AI agents across an organisation, understand their activities and enforce security policies while they work. The company is already working with enterprise customers including Micron, Peloton, ClickUp and Wix. Rather than attempting to stop companies from using autonomous AI, Bay wants to provide the security infrastructure that allows them to adopt it without surrendering control of sensitive systems and data.

Why AI Agents Are Creating a New Enterprise Security Problem
Traditional cybersecurity systems were largely designed around people logging into computers, applications and networks. AI agents change that operating model because they can independently read files, interact with applications, access cloud resources, execute code and perform tasks on behalf of employees. They may also continue working for extended periods without direct human supervision. The risk is not limited to hackers compromising an AI system. An agent can have legitimate access and still make an unsafe decision while attempting to complete an assigned task. For example, it might change a cloud configuration, expose a resource to the internet or transfer information through an inappropriate tool because it believes that action is necessary to finish its work. Bay says the idea for its platform emerged after its founders observed a coding agent attempting to expose a cloud environment while completing an internal proof of concept. There was no attacker involved; the agent was simply pursuing its objective in a dangerous way. This illustrates a fundamental problem with agentic AI: security systems must understand not only who or what has access, but also what an autonomous system is attempting to do with that access.
Bay Builds a Security Control Layer Across Enterprise AI Systems
Bay is developing a unified platform that discovers and governs AI activity across employee devices, browsers, cloud infrastructure and software-as-a-service applications. It aims to identify AI agents, coding assistants, local models, plugins, extensions, skills and Model Context Protocol servers, commonly called MCP servers. These components can give an agent access to external tools, business applications and internal information. Bay maps how they connect and what resources they can reach, helping security teams identify risky combinations and unexpected access paths. The platform also provides session-level visibility, following an agent from the original instruction through its decisions, tool calls and final actions. This broader view matters because a single command may look harmless when examined independently but become dangerous when considered alongside everything the agent has done before it. Bay’s objective is to give security teams one place to understand and control this expanding AI ecosystem rather than managing separate security policies for every model, assistant or agent framework.
Enforcer Evaluates Agent Actions Before They Are Executed
The most important part of Bay’s platform is its runtime enforcement system, called Enforcer. It evaluates proposed agent actions in the context of the full session, including the user’s original request, previous actions and the resources the agent is trying to access. Based on the organisation’s security policies, Bay can allow an action, ask for human approval, deny it or steer the agent towards a safer alternative. The company says these decisions are made locally in milliseconds, avoiding the need for a separate cloud round trip for every action. This approach is designed to catch behaviour that conventional rule-based systems may miss. For example, an agent accessing a file might be performing a normal business task in one situation but preparing to expose sensitive information in another. Looking at the broader task and activity can help distinguish the two. Bay also aims to address risks such as prompt injection and agent purpose drift, where an agent gradually moves away from its original instructions. The goal is to prevent dangerous actions without unnecessarily interrupting legitimate work.

Lighthouse Uses AI to Build and Adapt Security Policies
Enterprise security teams cannot realistically write individual rules for every new AI agent, model, plugin and workflow that employees introduce. Bay addresses this challenge through Lighthouse, its embedded AI-driven policy system. Lighthouse learns from an organisation’s historical agent activity and uses that context to recommend and maintain guardrails tailored to different teams. Rather than relying entirely on static rules, the platform attempts to understand how the organisation works, what resources different teams need and which activities present unacceptable risks. Bay says Lighthouse can analyse approximately 30 days of historical activity to help create policies before enforcement begins. The system also supports simulation, allowing security teams to examine what a proposed policy would allow, block or send for approval before it affects users. This is particularly relevant for large businesses where security settings must balance operational flexibility with strict controls. By making policies more adaptive, Bay hopes to reduce the administrative burden of securing an AI environment that may change every time employees adopt another agent or tool.
Enterprise Customers Need Visibility Without Slowing AI Adoption
Bay announced its emergence from stealth with dozens of enterprise customers, including Micron, Peloton, ClickUp and Wix. Its launch materials also report that Fortune 100 customers using the platform have been able to triple their autonomous AI agent usage while maintaining organisational control. That adoption figure is a company-reported claim, not an independently audited measurement, but it points towards the commercial problem Bay wants to solve. Enterprise leaders increasingly want employees to use AI for coding, research, customer operations and internal workflows, yet security teams need to understand where agents operate and what they can access. Blocking all AI activity can slow productivity, while allowing unrestricted access can create serious exposure. Bay’s proposition is to create an enforcement layer between autonomous agents and the systems they interact with. Its reported customer base suggests that this is becoming an operational requirement for some large organisations, rather than a theoretical concern limited to AI research teams.
The M Funding Gives Bay Resources to Expand Its Security Platform
Bay says the funding will support further product development, expansion of its engineering organisation and commercial growth across the United States and Europe. The distinction between the two financing stages is important: $13 million came from the earlier seed round led by YL Ventures and Striker Venture Partners, while the $37 million Series A was led by Battery Ventures with participation from Micron Ventures, M12 and Assaf Rappaport. The company is still at an early stage, having been founded in 2026, but its funding and customer announcements give it resources to move quickly into the emerging AI security market. Its founders bring experience from cybersecurity and technology organisations, including CrowdStrike and Noname Security, as well as Israeli military intelligence backgrounds. As enterprise agent deployments expand, Bay will need to support more tools, operating environments and complex workflows while maintaining reliable enforcement. It will also have to demonstrate that its platform can integrate into existing enterprise security processes without creating additional operational friction.
Bay Is Betting That AI Agent Security Will Become Essential Enterprise Infrastructure
Bay’s emergence from stealth reflects a larger shift in cybersecurity: the security perimeter is expanding from employees, devices and applications to autonomous software systems that can take actions on behalf of people. As AI agents gain access to business-critical systems, organisations need more than visibility into which models are being used. They need to understand the intent behind each task, the permissions available to an agent and the consequences of its next action. Bay is building around that requirement with a platform combining discovery, contextual monitoring, adaptive policy creation and real-time enforcement. Its opportunity is significant, but competition is growing as cybersecurity companies and enterprise software providers develop their own AI governance and agent protection capabilities. Bay must prove that its contextual enforcement works reliably across diverse environments and that organisations can trust it with critical workflows. The $50 million in total funding gives the company a substantial early foundation. Its longer-term success will depend on whether it can become a trusted security control layer for enterprises moving from AI experimentation to large-scale autonomous operations.
Sources
- Bay – Official $50M stealth launch and funding announcement, November 10, 2026
- SecurityWeek – Bay’s $50M funding, founders and AI agent security architecture
- Bay – Platform overview, AI discovery, Lighthouse and runtime enforcement
- Bay – Agent runtime enforcement and contextual guardrails
- Bay – Founders, company background and investors
- Reuters – Israeli AI security startup Bay raises $50M