Key Takeaways
- • Onyx Security announced a $113M Series B funding round led by Bessemer Venture Partners, positioning the startup to build what investors believe will become the dominant cybersecurity platform for enterprise AI agents.
- • The company has developed a secure control plane that discovers, monitors, secures, and remediates AI agents across an organization's entire technology footprint, including SaaS applications, cloud infrastructure, and endpoints.
- • Bessemer's thesis posits that AI agents represent a faster technological shift than cloud, endpoint, or network transitions, with adoption accelerating from <5% of enterprise work one year ago to a projected inversion where agents handle the majority of tasks within two years.
- • Onyx uses a proprietary mesh of small, purpose-built machine learning models that learn agent behavior patterns and operate inline, distinguishing its approach from legacy security architectures designed for human users and static devices rather than autonomous, high-privilege actors.
- • The agentic era security market is characterized by unique risk: agents inherit broad standing access by default and act with the credentials of systems they integrate with, yet most enterprises lack visibility into what individual agents are permitted to do.
Summary
Onyx Security, a startup building infrastructure for securing autonomous AI agents in enterprise environments, has closed a $113M Series B funding round led by Bessemer Venture Partners. The funding underscores venture capital's conviction that AI agents will become the defining technology platform of the next decade—and that the company controlling the security layer for this shift could exceed the scale of landmark cybersecurity predecessors including Palo Alto Networks, CrowdStrike, and Wiz.
The company's core product is a secure control plane—a unified platform that sits across an enterprise's entire AI footprint to discover, govern, and remediate AI agents wherever they operate. Unlike traditional security tools designed for human users or fixed devices, Onyx's platform addresses the unique challenge posed by autonomous agents: high-privilege actors that reason, call tools, move data, write and deploy code, and execute multi-step workflows independently at machine speed. The company's technical approach leverages a mesh of small, purpose-built machine learning models that learn how agents actually behave, allowing the platform to operate inline and detect anomalies or policy violations in real time.
What distinguishes the agentic era from prior technology transitions is both velocity and architectural risk. While cloud adoption unfolded over roughly a decade, agents are achieving enterprise penetration in quarters. Enterprise adoption has accelerated from less than 5% of work being handled by agents one year ago to a trajectory where that ratio inverts within two years—meaning swarms of agents performing the majority of work with human oversight on edge cases. This speed matters because agents inherit broad, standing access by default, acting with the credentials and system reach of their host infrastructure. Yet most enterprises currently lack visibility into what any given agent is actually permitted to do, creating exponential security exposure as deployment scales.
Bessemer's investment thesis frames this as a generational opportunity. The venture firm notes that each major technological shift—the internet, the endpoint, the cloud—produced one dominant security company. The agentic era, by this logic, will produce the largest cybersecurity company ever built, because agents represent a new category of actor that existing security architectures cannot adequately control. Onyx's control plane aims to fill that gap by providing security, governance, and infrastructure teams with unified visibility and enforcement across the entire agent ecosystem.
The company's core product is a secure control plane—a unified platform that sits across an enterprise's entire AI footprint to discover, govern, and remediate AI agents wherever they operate. Unlike traditional security tools designed for human users or fixed devices, Onyx's platform addresses the unique challenge posed by autonomous agents: high-privilege actors that reason, call tools, move data, write and deploy code, and execute multi-step workflows independently at machine speed. The company's technical approach leverages a mesh of small, purpose-built machine learning models that learn how agents actually behave, allowing the platform to operate inline and detect anomalies or policy violations in real time.
What distinguishes the agentic era from prior technology transitions is both velocity and architectural risk. While cloud adoption unfolded over roughly a decade, agents are achieving enterprise penetration in quarters. Enterprise adoption has accelerated from less than 5% of work being handled by agents one year ago to a trajectory where that ratio inverts within two years—meaning swarms of agents performing the majority of work with human oversight on edge cases. This speed matters because agents inherit broad, standing access by default, acting with the credentials and system reach of their host infrastructure. Yet most enterprises currently lack visibility into what any given agent is actually permitted to do, creating exponential security exposure as deployment scales.
Bessemer's investment thesis frames this as a generational opportunity. The venture firm notes that each major technological shift—the internet, the endpoint, the cloud—produced one dominant security company. The agentic era, by this logic, will produce the largest cybersecurity company ever built, because agents represent a new category of actor that existing security architectures cannot adequately control. Onyx's control plane aims to fill that gap by providing security, governance, and infrastructure teams with unified visibility and enforcement across the entire agent ecosystem.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.