Onyx Security raised $113 million in a Series B led by Bessemer Venture Partners, bringing total funding since its founding two years ago to $153 million. Cyberstarts, TCV, Conviction, FirstMark, Vintage, QuantumLight and G Squared also participated.
The company positions its product as a control plane for enterprise AI agents. It is designed to discover agent use across corporate environments, monitor actions at runtime and allow security, governance and infrastructure teams to stop unintended or malicious behavior.
Onyx said revenue has quadrupled since it emerged from stealth four months ago. The statement did not disclose absolute revenue, valuation or customer-contract values, so the growth multiple indicates acceleration but does not establish the company’s financial scale.
The platform targets a new operating problem: employees are granting increasingly capable agents access to critical systems and delegating complex tasks. Companies need visibility over third-party tools as well as agents built by internal engineering teams.
Onyx cited adoption by dozens of enterprises, including Fortune 500 companies in banking, technology, insurance and energy. It also highlighted a June integration with Anthropic and an implementation at Revolut, while keeping most customer identities and deployment volumes undisclosed.
The financing reflects investor demand for a security category built around agent accountability. Traditional access controls govern users and applications; Onyx argues that autonomous systems require monitoring of reasoning steps, policy compliance and intervention during execution.
The new capital will train the next generation of proprietary models and expand enterprise go-to-market operations in the United States and internationally. Execution will require converting pilots and integrations into durable contracts while keeping pace with rapidly changing model capabilities.
The next proof point is measurable commercial performance after the funding. Onyx must demonstrate retention, deployment depth and effective threat prevention, and it must substantiate that its control layer works across competing AI providers rather than becoming dependent on a narrow integration set.
