Onyx Security $113M Series B Targets AI Agent Governance

A big bet on keeping agents on a leash

Onyx Security has closed a $113 million Series B round to build out what it calls the Secure AI Control Plane, a product aimed at helping large organisations govern the AI agents that are increasingly running inside their systems 1. Bessemer Venture Partners led the round. Cyberstarts, TCV, Conviction, FirstMark, Vintage Investment Partners, QuantumLight and G Squared also participated 1.

The new money brings Onyx's total funding to $153 million 2. That implies roughly $40 million was raised in earlier rounds. The Series B is by far the larger tranche, which suggests investors see the company moving from early validation toward scaling.

What Onyx says it does

The company's pitch is simple to state and hard to deliver. As enterprises deploy more autonomous AI agents, humans need to stay in control of what those agents can see, do and decide 1. Reports on the round frame the product around governance, meaning visibility and policy enforcement over agents operating across a company's environment 12.

The "control plane" label is borrowed from networking and cloud infrastructure, where it describes the layer that sets rules and directs traffic rather than doing the work itself. Applied to AI, the term points to a central place where security teams can define what agents are allowed to do and monitor whether they stay within those limits. Neither report details the product's technical architecture. Claims about how it works beyond this broad framing should be treated cautiously until the company or its customers share more.

Why the timing matters

The round lands at a moment when AI agents are moving from demos into production. A chatbot answers questions. An agent can take actions on its own: calling APIs, reading internal documents, triggering workflows, sometimes chaining several steps together with little human review. Each new capability widens what security teams call the attack surface, and it makes accountability harder when something goes wrong.

Both outlets describe the core problem in similar terms: a growing population of agents inside enterprises that need oversight 12. The emphasis differs slightly. One account stresses the human-in-control angle 1. The other treats it more plainly as an enterprise governance and security story 2. These are two framings of the same product thesis rather than conflicting accounts.

Reading the investor list

The syndicate says something about how the market views this category. Bessemer is a long-standing enterprise and cloud investor, and Cyberstarts is known for backing security startups 1. TCV and G Squared typically write growth-stage checks. Conviction is closely associated with AI-native companies. Together they span cybersecurity, AI and later-stage software, which fits a company that sits where all three meet.

A nine-figure Series B also shows that investors are willing to fund AI security as a standalone category rather than waiting for incumbent security vendors to bolt on agent controls. Whether that bet pays off depends on whether "agent governance" becomes a budget line of its own or gets absorbed into existing identity, data security and cloud security platforms.

Open questions

The coverage leaves several things unaddressed. Neither report gives customer counts, revenue figures or a valuation 12. Neither says how Onyx plans to split the new capital between hiring, product development and expansion. And neither addresses competition. The field of startups and established vendors pitching AI security and agent oversight is crowded and getting more so.

There is also a definitional question the whole category has to answer. "Keeping humans in control" can mean anything from approving each sensitive action to simply having an audit trail after the fact. How much friction enterprises will accept in exchange for safety will shape which products win.

The takeaway

The clearest signal from this round is not about Onyx alone. Investors are now pricing AI agent security as a serious, venture-scale problem. Enterprises are clearly worried about autonomous software acting on their behalf, and that worry is generating demand. A $113 million round led by a top-tier firm 1, bringing total funding to $153 million 2, gives Onyx room to compete. The harder test is whether it can turn that capital into a product that security teams trust and that does not slow down the AI deployments their companies are racing to ship.