Rein Security raises $25M: runtime security for the AI agents enterprises actually build
Rein Security raised a $25 million Series A to secure the AI agents enterprises build in-house — runtime "sidecar" visibility into every line of code an agent executes, with real-time guardrails that block harmful actions before they complete.
Israeli startup Rein Security has raised $25 million in Series A funding to secure the AI agents enterprises are deploying faster than they can govern. The company announced the round this morning via PR Newswire, billing itself as "the AI-native runtime security platform for enterprises."
Two problems at once
The round was co-led by Glilot Capital and Sienna Venture Capital, with participation from Corner Ventures, Atlacle and RNP Capital Advisors, bringing Rein's total funding to $35 million, Calcalistech reported. Founded in 2024 by Matan Bar-Efrat and Netanel Rubin, veterans of Unit 8200, the company employs 31 people across offices in Tel Aviv and New York.
Rein's pitch targets two sides of the same problem: securing the AI agents a company builds for itself, and defending against increasingly sophisticated attacks powered by AI. Traditional security tools, designed around static perimeters and known vulnerabilities, were not built for systems that execute actions automatically — often on instructions received from another agent or system rather than directly from a human.
A sidecar, not a scanner
The technical differentiator is what Rein calls patented "sidecar" technology: instead of auditing an agent's code before or after deployment, the sidecar operates at runtime, while the agent is actually executing actions. That gives security teams visibility into every line of code an agent runs and every resource it touches, plus governance over agent activity and real-time guardrails designed to block harmful actions before they complete.
Two design choices stand out. First, Rein secures the AI agent supply chain without routing company or customer data through a gateway or proxy — a posture that avoids the data-exposure and latency concerns that come with funneling traffic through a middlebox. Second, the platform runs inside the customer's own cloud environment, a deliberate contrast with tools focused on agents built by outside vendors.
"We operate within the customer's cloud environment using a completely different technology," co-founder Netanel Rubin told Calcalistech. The focus is explicit: not the agents employees download from elsewhere, but the ones built internally — plus the shadow population of agents "developed independently on individual computers" that IT never approved.
Traction and a Black Hat demo
Rein is already generating revenue: since launching its product in January 2026, revenue is up eightfold and its customer base has grown fivefold, with the platform currently securing thousands of AI agents that collectively execute millions of actions, according to the company.
The company's vulnerability research team, Agent Breakers, presented at Black Hat USA 2026, demonstrating how it compromised the AI shopping agent of a top-five U.S. retailer — a reminder that adoption keeps outpacing the defenses. In another case Rein describes, an AI agent used by a global enterprise opened an ordinary-looking PDF that contained a hidden prompt injection instructing the agent to act outside its intended role; Rein says its technology detected and blocked the action before any damage occurred.
Why this round matters
The crowded agent-security field — Palo Alto's Nikesh Arora recently counted 40 companies racing to secure AI agents, at least 20 of them Israeli — is converging on a shared insight: agents need protection at execution time, not just at build time. Rein's bet is that the winning architecture watches from inside the customer's own cloud, sees every line an agent runs, and stops the bad action mid-flight rather than filing a report afterward. With agents now operating business systems at machine speed, that distinction may be the whole game.
Sources: Rein Security press release via PR Newswire (Oct. 8, 2026); Calcalistech (Oct. 8, 2026).