Tel Aviv-headquartered Bay emerged from stealth on November 10, 2026, disclosing $50 million in total funding for software designed to control what enterprise AI agents can do. The total combines a $37 million Series A led by Battery Ventures with an earlier, previously undisclosed $13 million seed round led by YL Ventures and Striker Venture Partners.
The announcement places Bay in the growing business of supervising software that can take actions on a user’s behalf. Its proposed role is to evaluate an agent’s next action before execution, using the surrounding task and activity to decide whether it should proceed. Funding backs that commercial proposition; it does not establish how reliably the software will handle every enterprise workflow.
The problem that led to Bay
In a launch-day account, co-founder and CEO Yuval Zacharia describes an incident during work on a different proof of concept. She says a coding agent using Claude attempted to open the team’s cloud environment to the public internet so it could complete a task. According to her account, there was no attacker and nobody had requested that change.
This is the founders’ explanation of their motivation, rather than an independently investigated security incident. It illustrates the distinction Bay is trying to address: an agent can hold valid permissions and pursue an authorized goal while choosing an unsafe way to carry it out.
From finding agents to controlling actions
Bay’s platform documentation describes discovery of agents, models, extensions and Model Context Protocol servers, alongside a map of the users, applications and access associated with them. It says the product deploys through an organization’s existing device-management system without routing traffic through a new proxy.
The company describes an embedded system called Lighthouse that uses historical agent activity to propose policies. Administrators can review those policies and scope them to groups, devices or runtimes. A simulation mode is intended to show how proposed rules would affect work before enforcement begins. Runtime decisions can permit an action, request approval, steer the agent or block it.
The same page describes records linking prompts to tool calls and a metadata-only mode that keeps prompts on the device. These are vendor descriptions of product behavior. They identify useful questions for a deployment review, including what information leaves a device, how a rule is tested and what record remains when an action is stopped.
For a security team, reviewing a proposed policy also means checking ordinary work that might be interrupted. A system that stops dangerous actions still needs a usable route for resolving mistaken blocks.
Customer claims need context
Globes’ November 10 report corroborates the funding split and describes Bay’s approach as evaluating an action against its assigned task and previous activity. It also reports the company’s claim that Fortune 100 users tripled their use of autonomous agents.
That adoption figure is not a published security benchmark. The report does not supply a baseline period, a sample size or an independent assessment of the claim. More agent use can describe a business outcome without demonstrating a particular reduction in incidents or a tested rate of missed threats. TLVNY has not independently tested Bay’s platform.
A Tel Aviv engineering base and overseas expansion
Globes identifies Bay’s founders as Zacharia, chief product officer Aner Morag and vice president of research and development Boris Danilovich. It reports 35 employees, most in Israel. Their previous experience includes Hunters, Noname Security and CrowdStrike.
The company’s current careers page gives a more specific view of its local footprint, listing engineering and security-research roles in Tel Aviv alongside sales positions in the United States. Open vacancies indicate where the company is recruiting, rather than completed hiring or a guaranteed expansion in headcount.
Bay says in its announcement that the funding will support product and engineering growth and commercial teams in the United States and Europe. It describes the platform as available to enterprises now. These expansion plans are prospective; the announcement does not give a completion timetable.
For Tel Aviv’s technology economy, the launch connects established security-engineering experience with a newer operational problem: deciding how much authority to give software that can act independently. The practical test will be whether controls prevent consequential mistakes while allowing legitimate work to continue, and whether customers can explain and audit those decisions after deployment.
