The list. Neo made the 2026–2027 Fortune Cyber 60 in the early-stage group. Fortune’s profile puts it in Boston with CEO Nick Warner, founded in 2025, 110 employees. Fortune reports $100 million in funding, from a16z, Bessemer, Craft Ventures and Merlin Ventures. That is the money Neo announced when it left stealth on July 20, 2026. The Wall Street Journal called it a combined seed and Series A. a16z says it led the seed. Bessemer says it led the Series A. Valuation and revenue are Undisclosed. Lightspeed, which helps run the Cyber 60, is not listed as an investor.
The team is the first reason this got funded. Warner ran go-to-market at SentinelOne and, as COO, took it public in 2021. Bessemer says he helped take it from $5 million to over $500 million in ARR. Co-founder Shlomi Salem spent more than a decade on SentinelOne’s detection and threat research. CTO Eran Shirazi co-founded EasySend and led vulnerability research in Unit 8200.
The problem in plain words. A developer installs an MCP server, a small plug-in that lets an AI coding agent reach other tools, with one command. A colleague adds a browser extension that promises AI summaries. Someone copies an agent “skill” from GitHub into a config folder. Nobody reviews any of it. Marketplace stars tell you an app is popular, not what it does. And Neo makes a sharp point: the app you approved last month is not always the app running today. One update can quietly give a familiar extension the power to read files.
The one unique thing: Neoverse. Most tools in this space build an inventory. Neo built a research engine. Neoverse crawls the places software comes from: VS Code, JetBrains, Cursor and Windsurf marketplaces, GitHub and the MCP registry, the Chrome, Edge and Firefox stores, and package managers like Homebrew and Winget. Neo’s endpoint sensor adds apps it finds on customer machines that never hit a public store. Then AI research agents check each app three ways. Static analysis reads the code, manifests and instruction files. Runtime analysis runs it in a sandbox and watches. Documentation analysis checks whether it does what it claims. Out comes a verdict, Malicious, Unsafe or Safe, re-run on every new version.
The verdict is the product. When Neoverse marks a skill malicious, Neo says it blocks it on any customer endpoint where it shows up, with no one writing a rule. Today that covers skills and MCP servers. Neo said Neoverse mapped more than 1.2 million pieces of software at launch. Its November 9 post says the public catalog now holds 4 million+ apps. Both are company figures.
The proof point. On November 10, Neo said Neoverse read the stripped Zoom desktop binary, with no source code, and found it was built with pjproject 2.14.1. That library carried CVE-2026-25994, a critical (CVSS 9.8) stack overflow, about seven months after the upstream fix. Neo’s human researchers then showed a remote crash with the return address under attacker control on macOS, Windows, iOS and Android. Neo says Zoom patched within 25 days; update to 7.2.0 or later. That is a real research team using AI to do real reverse engineering, not a demo.
Why it is winning now. Neo’s Cyber 60 release cites the Lightspeed Cyber60 Survey: AI agents are in production at 77% of surveyed companies, but only 20% report a complete agent inventory. Neo’s own October survey of 50 CISOs and CIOs found 86% have AI in production, yet only 23% say their tools can stop a misbehaving agent mid-action. Small sample, company-run. The direction still matches what buyers say.
The CyberMerge view. Discovery is turning into a commodity. This series has covered JetStream mapping agents, Keycard issuing agent tokens and Ent judging intent on the endpoint. Every one of them can “find your agents.” Neo’s own survey says discovery is what current tools already do best. So the inventory is not the moat.
The research is. This is the old antivirus reputation playbook, a cloud brain that knows every file, rebuilt for MCP servers, skills and extensions. SentinelOne alumni know exactly how that game is won. Giving Neoverse away free behind an email sign-in is smart: it makes the brain famous before the sensor is.
The risks are familiar too. Neo needs a sensor on the laptop, which means fighting for a slot next to CrowdStrike or Microsoft. And a reputation database is only worth paying for if it stays ahead of the attackers and the incumbents.
Bottom line. Neo is betting that “what is this software?” is a better question than “what is this agent doing right now?” Watch for named customers, a published false-verdict rate, and blocking that reaches beyond skills and MCP servers. If Neoverse becomes the place defenders check before they install, Neo owns a category. If not, it is one more agent inventory.
