Onyx Security Raises $113 Million to Build a Control Layer for AI Agents

Onyx Security Raises $113 Million to Build a Control Layer for AI Agents

Four months ago, Onyx Security was a stealth-stage startup that most enterprise buyers had never heard of. On July 29, it announced a $113 million Series B led by Bessemer Venture Partners, a round that values the two-year-old Israeli company at an estimated $640 million and vaults it into the front rank of a category that barely existed a year ago: security for autonomous AI agents.

The raise brings Onyx's total funding to $153 million, following a $5 million seed, a $35 million Series A, and the combined $40 million the company disclosed when it emerged from stealth this spring. Bessemer was joined by Cyberstarts, TCV, Conviction, FirstMark, Vintage, QuantumLight, and G Squared. The company said the money will go toward training its proprietary security models and expanding sales, business development, and marketing in the United States and other international markets.

What Onyx does

Onyx pitches itself as a dedicated "control layer" for AI. As enterprises hand agents access to SaaS applications, cloud infrastructure, and endpoints, those agents increasingly take consequential actions with little built-in accountability or oversight. Onyx's platform is designed to let organizations discover where AI is being used, including shadow deployments, manage what agents are permitted to do, and monitor their behavior in real time.

The technical core is a set of proprietary models that track an agent's reasoning at every step. If an agent attempts an unauthorized, unexpected, or malicious action, Onyx says the system intervenes to stop or correct it before damage is done. The platform is built to enforce real-time safeguards against threats such as prompt injection and to help enterprises meet regulatory compliance requirements. It governs agents whether they are built internally or sourced from third parties.

The company says it already secures more than 1.1 million agents across enterprise customers and inspects more than 66 million AI sessions in real time. It counts several Fortune 500 companies among its customers, and in June, Anthropic announced it had integrated Onyx's technology to help its enterprise customers adopt AI more securely. Onyx says revenue has quadrupled since it left stealth, growth that co-founder and CEO Maxim Bar Kogan credits for pulling the Series B forward.

"Coming out of stealth marked the point at which we could begin selling broadly," Bar Kogan told Calcalist. "Demand exceeded our expectations, we reached our targets very quickly, and that created tremendous investor interest and allowed us to raise this round much sooner than expected."

The founders

Onyx was founded in 2024 by Bar Kogan and Gil Elbaz. Bar Kogan is a veteran of Unit 8200, Israel's elite signals-intelligence corps, and a former vice president of product and engineering at Mixtiles. Elbaz is an AI entrepreneur who served in the Israeli Air Force's operational technology unit. The company now employs more than 80 people across Israel, the United States, and Canada.

Bar Kogan frames the raise as a bet on a market that does not yet have an established leader. "The amount raised reflects investors' belief that this category will produce very large companies and become one of the most important areas in cybersecurity," he said. "Our mission is to control AI. Trillions of dollars are being invested in AI, and the companies that help enterprises deploy it safely will be worth billions. Very few companies today can provide comprehensive control over AI agents."

Why agent security is 2026's hot category

Onyx's round lands in the middle of a funding wave aimed squarely at the risks created by agentic AI. As models graduated from chat assistants to systems that can browse, transact, write code, and act on behalf of employees, security teams found themselves without tools built for machines that make their own decisions. Investors have responded by pouring capital into a cluster of startups building governance, identity, and monitoring for AI, alongside a parallel run of acquisitions as incumbents scramble to catch up.

The thesis is straightforward: every enterprise deploying agents needs a way to see them, constrain them, and prove they behaved. Hila Zigman, a general partner at Cyberstarts, which led Onyx's seed round, argues the problem is large enough to define a new market.

"As AI agents become an integral part of critical enterprise workflows, organizations need to ensure they operate securely, consistently, and in accordance with corporate policies," Zigman said. "Onyx Security is building the control layer that will enable enterprises to adopt AI at scale. We believe this will become one of the most important cybersecurity categories of the coming decade."

Bar Kogan sees the same shift scrambling the competitive order. "AI is reshuffling the cards across every industry, including cybersecurity. Every incumbent is vulnerable, and smaller companies have an opportunity to redefine the market," he said.

What to watch

The open question is whether "control layer for AI" becomes a durable standalone category or gets absorbed into the platforms of larger security vendors, several of which have been acquiring their way into the space. Onyx says it intends to build an independent, global company rather than sell, and the $640 million valuation gives it room to press that ambition. Watch three things next: whether the Anthropic integration is followed by deals with other frontier model providers, whether Onyx can convert Fortune 500 pilots into large recurring contracts as its revenue base matures beyond its first year of broad selling, and whether the proprietary security models it plans to fund with this round can keep pace as the agents they police grow more capable. In a market this young, the lead can change hands quickly.