Neo Security

**Product and the concrete problem it solves.** Neo Security (branded simply "Neo," at neo.ai) is attacking a control gap that has opened almost overnight as enterprises deploy autonomous and semi-autonomous AI software. The concrete problem: AI agents, copilots, Model Context Protocol (MCP) servers, browser extensions, and "skills" now take real actions on real endpoints — reading files, calling APIs, moving data, invoking tools, and chaining workflows — but most security stacks were built to govern *humans* and *static applications*, not autonomous non-human actors whose behavior is probabilistic and whose permissions are frequently over-broad. Neo's stated mission is "agentic software control," captured in its tagline: "Reveal, understand, and control every human and non-human action on every endpoint in real time." In practice, Neo answers three operational questions for a security team: (1) *what* agentic and traditional software is actually running (agents, models, extensions, MCP servers, skills); (2) *how* it is configured, flagging excessive access privileges and misconfigurations; and (3) *what happens when it runs*, enforcing policy at the moment of action. The pitch is that you cannot secure autonomous software you cannot see, and you cannot govern it with tools that only observe posture after the fact.

**Core technology and how it actually works.** Neo's architecture is deliberately endpoint-centric rather than cloud-log-centric, which is its most important technical bet. The company describes an on-device agent that performs local analysis with "no kernel module, no reboot" — a lightweight deployment posture that lowers adoption friction and keeps sensitive analysis on the endpoint for data-privacy reasons. At the core is **Neoverse**, a continuously updated knowledge base that maps the capabilities, risks, and behavioral patterns of agentic software so that a novel agent or MCP server can be characterized rather than treated as an opaque process. The enforcement mechanism is the differentiator: Neo says it *intercepts* an agent's actions at the endpoint and can **allow, block, or hold for approval** before the action reaches sensitive data or systems — a real-time control point, not merely a dashboard. Around that sit a continuous catalog/inventory of active elements (agents, models, extensions, MCP servers), **real-time attribution** that traces a given software action back to a specific human user or application, granular policy enforcement over tool calls, data movement, agentic workflows, and API access, and audit trails that record both human and AI actions. The combination — discover, characterize via Neoverse, attribute, and enforce inline — is what Neo means by a "control layer" as opposed to the discovery-and-posture approach common in early AI-security tooling.

**Market, customers, and go-to-market.** Neo sells into enterprise security operations teams — the same buyers who already run endpoint detection and response (EDR), identity, and data-security programs — and positions "agentic software control" as a new adjacent category rather than a feature of an existing one. The go-to-market thesis rides a genuine wave: as of 2026 enterprises are rapidly piloting AI agents and MCP-based integrations, and CISOs are visibly anxious about non-human identities, agent sprawl, and the blast radius of an over-permissioned autonomous workflow. Neo's endpoint-first, privacy-preserving deployment ("no kernel module, no reboot," local analysis) is tuned to reduce the procurement and rollout friction that slows security-tool adoption. The company emerged from stealth on July 20, 2026, so named marquee customers, pricing, and revenue are not yet public and should be treated as unconfirmed; the go-to-market motion (enterprise direct, security-operations buyer, land-and-expand from endpoint visibility toward inline enforcement) is inferable from the product and team but not yet evidenced by disclosed logos.

**Traction, funding, and third-party validation.** Neo's strongest external validation is the caliber of its capital and backers relative to its age. The company launched from stealth with **$100 million in total funding** — a **$25 million seed round completed in 2025** and a **~$75 million Series A** — led by **Andreessen Horowitz (a16z)** and **Bessemer Venture Partners**, with participation from **Craft Ventures** and **Merlin Ventures**. Reporting also lists a notable angel roster including **Wiz CEO Assaf Rappaport** and former Wiz executive **Merav Bahat**, among others — an unusually strong cybersecurity-operator syndicate for a company at emergence. As of the July 2026 launch, Neo reported roughly **50 employees, about 40 of them in Israel**, indicating an Israel-centered R&D base with a U.S. commercial footprint. What remains unverified and should be diligenced directly: paying customers, ARR, retention, the depth and independence of the Neoverse knowledge base, and any third-party security evaluations or design partners. In short, the validation here is investor- and operator-driven rather than customer-proven — appropriate for a stealth-exit stage, but not yet market proof.

**Founders and team background.** Neo's team is its most defensible asset. CEO **Nick Warner** was previously **President and COO of SentinelOne**, where he built the go-to-market organization and helped carry the company through its 2021 IPO, with earlier senior roles at **Cylance, McAfee, and Forcepoint** — a rare profile of someone who has scaled an endpoint-security business from growth stage to public-company scale. CPO **Shlomi Salem** led **SentinelOne's detection engineering for roughly a decade** and co-led internal threat research, giving Neo deep endpoint-detection and adversary-behavior expertise precisely where the product must be strong. CTO **Eran Shirazi** co-founded the software company **EasySend** and, per reporting, previously **led a vulnerability-research group in the IDF's Unit 8200** — the elite signals-intelligence unit that is a canonical source of Israeli cyber founders. The pairing of a proven endpoint-security operator (Warner), a career detection engineer (Salem), and an offensive-research and product builder (Shirazi) is well-matched to "control every action on every endpoint." The principal open question is organizational maturity beyond the founding trio — sales leadership, customer-success, and channel — which is normal for a company weeks out of stealth.

**Competitive dynamics.** Neo is entering one of the most contested arenas in security, where incumbents and startups are converging on "AI/agent security" from multiple directions. (1) **Endpoint incumbents** — CrowdStrike and Neo's own alma mater SentinelOne — can extend EDR toward agentic control and enjoy installed-base distribution. (2) **Cloud/AI-security platforms** — Wiz (whose CEO is a Neo angel), Palo Alto Networks (Prisma AIRS), and Microsoft — are folding AI and agent security into broad suites. (3) **Non-human / machine-identity players** — CyberArk, Okta, Silverfort, and AuthMind — attack the identity-and-privilege side of the same problem. (4) **A wave of agent/LLM-security startups** — including Noma Security, Zenity, Aim Security, Prompt Security, and Lasso Security — compete directly on securing agents, MCP, and LLM workflows, several with a head start. Neo's plausible edges are: (i) **inline endpoint enforcement** (allow/block/hold at the moment of action) rather than discovery-and-posture only; (ii) an **on-device, privacy-preserving, no-kernel-module** deployment that eases adoption; (iii) the **Neoverse** behavioral knowledge base as a data moat if it compounds; and (iv) a founding team that has already built and scaled category-defining endpoint software. The countervailing risk is that "agent security" is consolidating fast and the largest incumbents can bundle a "good enough" capability into platforms customers already own.

**Defense, security, and resilience dual-use relevance.** Neo's dual-use relevance should be stated with calibration: it is an enterprise cybersecurity company, and its defense/resilience weight is an adjacency, not a fielded government capability. The credible thesis is that governing autonomous AI agents and non-human identities is becoming a **critical-infrastructure and national-security resilience problem**: as government agencies, defense organizations, and operators of energy, finance, and telecom infrastructure adopt agentic AI into workflows and even command-support systems, the ability to inventory, attribute, and inline-control every autonomous action — and to pause or block a compromised or misbehaving agent before it touches sensitive systems — maps directly onto the emerging attack surface those institutions face. The Unit 8200 pedigree on the founding team and the endpoint-enforcement design are consistent with security-grade requirements. Honest calibration: there is no public evidence of defense contracts, government deployments, or accreditations (e.g., FedRAMP/IL), and the roadmap is enterprise-first; dual-use here is a real and strategically-timely adjacency that would only become a fielded capability with government customers, accreditation, and hardened deployment that the public record does not yet show.

**Growth stage, trajectory, and key diligence risks.** Neo reads as an **early-stage** company with an unusually strong launch: founded in 2025, emerged from stealth in July 2026 with $100M and elite backers, product publicly described but not yet market-validated. The trajectory is promising precisely because the team has done this before, but the risks are concentrated and real. (1) **Category crowding and incumbency:** agent security is filling with well-funded startups and platform incumbents (CrowdStrike, SentinelOne, Wiz, Palo Alto Networks, Microsoft) who can bundle. (2) **Product-maturity risk:** inline endpoint interception of arbitrary agentic actions is technically hard to do reliably without breaking legitimate workflows, and Neoverse's value depends on breadth and accuracy that must be earned over time. (3) **Metric opacity:** customers, revenue, retention, precise HQ, and evaluation results are undisclosed. (4) **Expectation and burn risk:** a $100M launch sets a high bar and a fast clock. (5) **Dual-use is adjacency, not fielded**, so the strategic-defense case is prospective. Progression signals to track: named enterprise customers and design partners, disclosed ARR and net retention, independent security evaluations of the enforcement layer, MCP/agent-ecosystem partnerships, and any move toward government accreditation that would convert the resilience thesis into a fielded capability.