Bay Raises $37M Series A for Enterprise AI Agent Security

Before a security team applies a new rule to an AI agent, Bay offers to show what that rule would have done to the previous 30 days of work. That rehearsal puts an operational question beside the Tel Aviv company's financing: how should an enterprise decide which automated actions deserve an interruption?

Bay disclosed a $37M Series A led by Battery Ventures on November 10, 2026, together with a previously undisclosed $13M seed led by YL Ventures and Striker Venture Partners. The $50M total supports an AI agent security platform built to examine proposed actions in the context of a session, giving security teams a way to carry policy into software that acts on their behalf.

Micron Ventures, M12, Microsoft's Venture Fund, and Wiz CEO Assaf Rappaport also participated in the Series A, according to Bay's funding announcement. The company plans to expand product and engineering, alongside sales, marketing and customer teams in the United States and Europe. The commercial proposition reaches beyond identifying an agent: someone has to decide what it may do next.

Bay's financing follows a problem with legitimate permissions

Bay's founders describe watching a coding agent attempt to expose a cloud environment to the internet while trying to complete a test. In the company's account, the agent was pursuing an assigned task with valid permissions. The risk emerged from the route it chose to finish the work, which helps explain why Bay puts the proposed action inside the security decision.

The official founder roster identifies Yuval Zacharia as CEO, Aner Morag as CPO and Boris Danilovich as VP of R&D. All three are co-founders. Their earlier work includes research and AI at Hunters, product at Noname Security and engineering at CrowdStrike, respectively, bringing endpoint and application-security experience to a product aimed at autonomous workflows.

Bay was founded in 2026 and is headquartered in Tel Aviv. Its financing separates two stages of the business: a $13M seed round that was disclosed alongside the new $37M Series A. Treating the combined $50M as one Series A would obscure the sequence of investment behind the company.

Battery's decision to lead follows the earlier backing from YL Ventures and Striker Venture Partners. YL Ventures' portfolio places Bay in AI agent security and confirms its Tel Aviv base and 2026 founding. The financing gives the team resources to develop a control product while building the customer-facing organization needed to support its use.

A policy has to understand the work it interrupts

On its platform page, Bay describes discovery of agents, MCP servers, extensions, models and skills, along with a map of the systems and credentials they can reach. The company also describes session-level records connecting prompts to tool calls. Those are product capabilities presented by Bay, rather than independently tested guarantees of complete visibility.

The combination matters because an enterprise can approve an AI tool while remaining uncertain about an individual operation. An agent's access describes what is technically available; the assigned task provides the reason for using it. A security team trying to evaluate an action needs enough of both to make a decision that fits the business.

Bay says its Lighthouse system uses an organization's agent activity to recommend policies and lets teams examine those policies against historical work. Its product description includes scoping by identity-provider group, device or runtime. A rule can therefore be evaluated against the way a team works before it begins generating requests for human approval.

That rehearsal changes the implementation conversation. Security staff can ask which actions a proposed rule would interrupt, where a person would need to step in and whether the available record explains why. Those questions connect product configuration to the people expected to operate it, with the consequences visible before enforcement becomes part of the working day.

Runtime enforcement creates a continuing customer relationship

Bay's runtime-enforcement documentation describes decisions made locally in milliseconds, with options to allow, ask, deny or steer an agent. It says the decision considers the whole session and can redirect work when activity drifts from policy. These remain company-reported design and performance claims; buyers need to evaluate them in their own environments.

A product operating at that point in a workflow inherits a demanding responsibility. Too many unnecessary interruptions can make an approved tool difficult to use, while a permissive policy can allow an action the organization meant to prevent. The useful evaluation concerns the quality of the decision and whether the security team can understand and adjust it as the work changes.

Bay names Micron, Peloton, ClickUp and Wix among its customers. Micron's participation as an investor connects the financing to a disclosed commercial relationship, although the announcement does not provide contract values or customer-level results. The breadth of named customers establishes who Bay says it serves without establishing identical deployments or outcomes across those organizations.

The company's short deployment and response-time claims belong in that same evaluation. Installation, policy calibration and ongoing investigation are different jobs. An enterprise considering Bay has reason to examine each one, including how local decisions are recorded, how human approval fits the existing security process and how policy changes affect established agent workflows.

The expansion plan carries security into everyday operations

Bay's careers page lists engineering and research positions in Tel Aviv and sales roles in the United States. That hiring footprint is consistent with the announced use of capital, which combines further product development with a larger commercial and customer organization. The company has not disclosed a valuation or a detailed hiring target in its funding announcement.

For security operators, the financing is attached to an expanding operating relationship. As employees add agents, connect tools and extend assignments, the organization needs its control decisions to remain understandable to the people responsible for them. Historical simulation offers a place to examine that relationship before the next rule goes into effect.

The founders are using the round to build both the software making those decisions and the teams supporting its customers. Inside a customer organization, the work continues with a more specific question: when an agent's next action needs a person, will that person have enough context to answer and let the rest of the assignment proceed?