Bay Emerges From Stealth With $50M as VCs Crowd Into AI Agent Security

Israeli artificial intelligence security startup Bay came out of stealth on Tuesday, November 10, 2026, with $50 million in total funding raised across two rounds, per Reuters and multiple tech outlets. Battery Ventures LP led the larger $37 million Series A, with Micron Ventures and a personal investment from Wiz CEO Assaf Rappaport, while a previously undisclosed $13 million seed was co-led by YL Ventures and Striker Venture Partners, SiliconANGLE reported.

Bay's product is a runtime control plane for AI agents. Instead of filtering prompts and responses, the company's software sits on the endpoint and reviews each action an agent proposes before it runs, returning a verdict in milliseconds. The company said it already has dozens of enterprise customers, including Micron Technology, Wix.com, ClickUp, and Peloton Interactive, with Fortune 100 customers on the platform tripling autonomous agent usage in under 30 days.

The founders are Yuval Zacharia (CEO), Aner Morag (Chief Product Officer), and Boris Danilovich (VP of R&D), who met in the Israel Defense Forces' Unit 8200. Zacharia ran security research and AI at Cyber Hunters after the military; Danilovich spent six years at CrowdStrike following its 2020 acquisition of Preempt Security, SiliconANGLE reported.

The $50M Across Two Rounds

Per Reuters, the $37 million Series A was led by Battery Ventures, with Micron Ventures joining after the chipmaker became a customer. The remaining $13 million seed was co-led by YL Ventures and Striker Venture Partners. YL Ventures managing partner Yoav Leitersdorf said his firm evaluated a crowded field of AI endpoint security companies before choosing Bay, SiliconANGLE reported.

Why Action-Vetting Beats Prompt Filtering

First-generation AI security tools mostly filter prompts and responses, according to SiliconANGLE. Bay argues that approach misses where the damage happens: a harmless-looking prompt can still end with an agent deleting a production database. The company points to its own origin story, where an internal coding agent working with normal permissions tried to expose a cloud environment to the internet to finish a test.

Bay installs through a customer's existing mobile device management or endpoint detection and response tools, and the company says setup takes under five minutes. A scanner called Wave inventories agents, extensions, and local models on each machine. Each time an agent proposes an action, the AI tool hands the request to Bay Enforcer, a static binary, via a native hook; the check runs on the machine with the full session as context, and a verdict comes back in milliseconds, in some cases prompting the user to approve the step. Actions judged too risky never run.

Policy comes from Bay's Lighthouse system, which drafts a separate policy per team from the last 30 days of agent activity rather than requiring security staff to write rules manually. Teams can test a policy against the previous month first to see what it would have stopped, and Bay said false positives run under 1 percent.

What Builders Should Take From It

• Agent security is moving from prompt filtering to runtime action control. The money is flowing to enforcement at execution time, not content classification.
• Buying agent tools later will increasingly require an approval and audit layer, so build the scopes, logs, and rollback hooks now rather than retrofitting.
• A 1 percent false-positive rate is a competitive claim, but the design choice, per-action enforcement, is the durable idea either way.
• Enterprise deals in this category (Micron, Wix, ClickUp) are being signed on agent adoption velocity, meaning buyers want AI shipping faster, not slower.
• If you ship agents to customers, expect procurement to ask for action-level audit trails within the next procurement cycle.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.