Neo Security exits stealth with $100 million for AI agent security

Neo Security said it has raised $100 million and is coming out of stealth with software meant to help security teams monitor and control AI agents inside enterprises. Andreessen Horowitz and Bessemer Venture Partners led the financing, with Craft Ventures and Merlin Ventures also participating, as investors back a new security category around agentic software.

The company did not disclose its valuation, revenue, customer count or headcount. It said the money will go toward expanding engineering and go-to-market teams.

Neo was founded by former executives and engineers from SentinelOne, Wiz and Palo Alto Networks. Its pitch is that security operations teams need a dedicated system for inventory, risk assessment, attribution and policy enforcement as AI features are added to browsers, developer tools, SaaS products and older enterprise applications.

The funding lands as vendors across enterprise software are adding agentic capabilities to products that already sit inside approved corporate environments. That creates a more awkward security problem than a new unsanctioned app: the agent may inherit a user’s permissions, call tools, move data or work through workflows in ways that appear valid to legacy controls.

What Neo says it is building

Neo describes its product as a real-time control layer for AI-enabled and agentic software. According to the company, the platform inventories AI agents, AI-enabled applications, plugins, extensions, MCP servers and traditional software that has embedded agentic functions.

The company says its system then maps what those tools can do, what they can access and how they are configured. It also says it provides attribution for agent actions, so security teams can review an audit trail tied to automated work rather than treating agent activity as an indistinct extension of a human user.

Neo also claims to support granular controls based on groups or identities, including policies for tool calls, API access, data movement and agentic workflows. The company said its enforcement features can block risky behavior and malicious models, though it did not provide customer metrics or independent benchmarks for those claims.

The category bet

Neo is leaning on a forecast from Gartner, which said 5% of enterprise applications had agentic features at the end of last year and more than 40% will have them by the end of 2026. If that forecast holds, the security problem will spread through existing software estates rather than remain confined to a handful of AI-native tools.

Nicholas Warner, Neo’s co-founder and chief executive, said existing enterprise security products were designed around more predictable software behavior. He said agentic capabilities now being embedded into common applications let software reason, act, invoke tools and operate with user permissions.

Andreessen Horowitz General Partner Zane Lackey said applications change materially once they gain agentic capabilities, including what they can access and what harm they can cause. He said organizations need visibility into the behavior of that software and controls that can govern it in real time.

The round is large for a company just leaving stealth, but the investor logic is straightforward: if agentic software becomes common inside enterprise applications, SecOps teams will need controls that sit closer to behavior and permissions than conventional application inventories. Neo still has to prove that this becomes a budgeted security category rather than a feature absorbed by incumbent security platforms.

This story draws on original reporting from SiliconANGLE.