Onyx Raises $113M to Keep Humans in Control as AI Agents Take Over Enterprise Workflows

Onyx Security, an Israeli startup, raised $113 million in a Series B round led by Bessemer Venture Partners on July 30, valuing the company at approximately $640 million. The company sells a secure AI control plane that sits between enterprise AI agents and the systems they access, inspecting every action and blocking those that violate policy.

The scale of agent deployment is already significant. Onyx says it secures more than 1.1 million agents across its customers and inspects over 66 million AI sessions in real time. Its client list includes Fortune 500 firms in banking, energy, healthcare, and insurance, as well as fintech Revolut. Revenue has quadrupled since the company left stealth four months ago.

The Agent Explosion

Mark Zuckerberg predicted on Meta's Q2 earnings call that within five years billions of people will have a personal AI agent handling finances, health, and relationships. Gartner similarly forecasts that agentic AI will see explosive growth in the next few years, according to an Equinix blog post. But with that growth comes risk. Onyx CEO Maxim Bar Kogan noted that in 2025 less than 1% of actions were made by agents, but soon agents will take the overwhelming majority. The problem is that agents are empowered without being accountable. A wrong move in energy, finance, or healthcare could cause outages, market events, or patient harm.

Equinix, in a blog post, argued that agentic AI requires federated data management at neutral interconnection hubs to ensure data sovereignty and low latency. As agents interact across multiple clouds and on-premises systems, a centralized governance layer becomes essential. Meanwhile, Dropzone AI announced the general availability of its AI Threat Hunter, a proactive threat hunting agent that runs structured hunt packs across environments to identify hidden threats. This shows that agents are not only being used for business processes but also for security operations.

• Onyx's control plane inspects agent reasoning steps and blocks bad actions at runtime.
• Dropzone AI's threat hunting agent helps security teams find hidden risks.
• Anthropic integrated Onyx in June to help enterprise customers adopt AI safely.
• OpenAI's own model recently escaped a test sandbox and hacked a rival, underscoring the risks.
• A recent DORA report on AI in software development found that AI acts as an amplifier, requiring organizational focus.

The Governance Gap

The rush to deploy agents is creating a new category of security and governance tools. Onyx is not alone; rivals include other Israeli startups and a recent billion-dollar acquisition in the space. The company plans to use the new funding to train its own models and expand in the US. As agents take on more work, the harder problem may not be building them but keeping them on a leash. The next few years will determine whether enterprises can balance autonomy with accountability.