Onyx Security has raised $113 million to develop technology intended to keep autonomous artificial intelligence agents under enterprise control as U.S. companies give AI systems broader access to software, sensitive data and operational workflows.

The Series B round was led by Bessemer Venture Partners, with participation from Cyberstarts, TCV, Conviction, FirstMark, Vintage Investment Partners, QuantumLight and G Squared. The financing brings Onyx’s total capital raised to $153 million since the company was founded in 2024. Onyx did not formally disclose its valuation, although The Times of Israel reported that the transaction valued the startup at approximately $640 million.

The size and speed of the round reflect growing investor interest in security products built specifically for agentic AI. Onyx emerged from stealth only four months ago with $40 million in previously announced funding. The company said its revenue has since increased fourfold, though it did not disclose revenue figures, customer spending or other financial details that would allow independent comparison.

Onyx was founded by Chief Executive Maxim Bar Kogan and Chief Technology Officer Gil Elbaz. Bar Kogan previously worked in Israel’s Unit 8200 cyber intelligence organization and held a product and engineering leadership role at Mixtiles. Elbaz previously worked as an AI architect at Nvidia and served in an Israeli Air Force technology unit. The company maintains offices in New York and Tel Aviv and has employees in Israel, the United States and Canada.

The startup is targeting a security problem that differs from the risks associated with ordinary generative-AI assistants. Autonomous agents can be authorized to browse websites, write and run code, retrieve corporate files, communicate with other applications and complete business processes without requiring a person to approve every step. As those systems receive more credentials and permissions, a mistaken or manipulated agent could potentially expose data, alter records or trigger actions across multiple services.

Traditional cybersecurity products often focus on human identities, devices, network traffic and individual applications. Onyx argues that enterprises need an additional control layer capable of identifying AI agents, observing their activity and deciding whether particular actions comply with company policy. The goal is to give security teams visibility into agents developed internally as well as third-party agents introduced through commercial AI products.

According to the company, its platform operates across browser, endpoint, software-as-a-service and cloud environments. It discovers agents active inside an organization, evaluates their actions before they take effect and can approve, redirect or block behavior considered dangerous or unauthorized. Human reviewers can be brought into the process when an automated decision requires escalation.

Onyx also markets defenses against prompt injection, a class of attack in which malicious instructions embedded in data, websites or documents attempt to manipulate an AI system. Such attacks are especially concerning when an agent can use external tools or has access to business systems. Other risks include agents obtaining excessive privileges, operating through unapproved applications or taking actions that are difficult for compliance teams to reconstruct later.

The company says it now secures more than 1.1 million agents and inspects over 66 million AI sessions in real time. It has described its customers as including Fortune 500 companies in banking, insurance, energy, technology and other industries, although it has not released a complete customer list. Revolut’s head of security engineering was quoted in the company’s funding announcement describing Onyx as providing visibility and runtime controls for coding agents used across multiple service providers.

Onyx also highlighted a June integration with Anthropic, the San Francisco AI company behind the Claude model family. Onyx says its platform connects with dozens of AI providers and lists integrations involving major U.S. technology companies, including OpenAI, Amazon, Microsoft and Google. Those connections are important because enterprises increasingly use models and agents from several vendors rather than relying on a single AI platform.

The new funding will be used to train additional proprietary models and expand enterprise sales and customer support in the United States and other markets. Onyx said it is prioritizing organizations in financial services, healthcare and energy, where an autonomous system’s error could have consequences beyond an isolated software failure.

For U.S. chief information security officers, the emerging challenge is not only whether an AI model produces inaccurate information. Agents can function as machine identities with continuing access to corporate resources. Security teams therefore must determine what each agent is allowed to see, which tools it may use, how its activity is recorded and when a person must intervene.

That requirement is creating a new market alongside identity security, cloud protection, data governance and application security. Established vendors and other startups are also introducing products that inventory AI systems, monitor agent behavior and enforce access policies. Onyx will have to demonstrate that its approach can work across rapidly changing models and agent frameworks without adding unacceptable delays or generating excessive false alarms.

Many of the company’s scale and performance figures remain self-reported, and the wider enterprise market is still determining how quickly autonomous agents will move from pilot projects into high-risk production systems. Large organizations may also favor security functions incorporated directly into existing cloud, identity or endpoint platforms rather than purchasing a separate control layer.

Still, the financing indicates that major venture investors believe agent governance could become a durable cybersecurity category. Bessemer partner Amit Karp said enterprises need trust, control and accountability before deploying agents broadly, describing Onyx as a potential control layer for that transition.

For Silicon Valley, the round illustrates a second-order investment opportunity created by the AI boom. As model developers race to make agents more capable and independent, security companies are raising substantial capital to ensure those systems remain observable, accountable and subject to human authority.