Neo Security, a cybersecurity startup founded by former SentinelOne executives, has emerged from stealth with $100 million in total funding—significantly more than the reported $50 million figure circulating earlier. The Series A round closed in July 2026 with $75 million in backing from heavyweight investors including Andreessen Horowitz and Bessemer Venture Partners, capping off a dramatic entry into one of cybersecurity’s most urgent frontiers: securing artificial intelligence. The company’s core insight reflects a shift in how security leaders think about AI deployment: autonomous AI agents operating within enterprise networks with legitimate permissions pose fundamentally different attack surfaces than traditional software vulnerabilities.
Led by Nicholas Warner, who served as President and COO of SentinelOne before founding Neo Security, the company brings seasoned expertise in endpoint security to an entirely new problem space. When a junior analyst asks an AI agent to summarize monthly budgets, that agent might legitimately access payroll systems, customer records, or financial databases. The risk isn’t that the AI is hacked—it’s that the AI’s own autonomous decisions, unmonitored and unchecked, could extract, modify, or misuse that data without anyone noticing. Neo Security’s product is designed to sit between those agents and enterprise data, creating a control layer that monitors and restricts what even authorized AI can access and do.
Table of Contents
- Why Former SentinelOne Leaders Are Betting on Enterprise AI Security
- Building a Control Layer for Enterprise AI at Scale
- The Investor Thesis and Series A Details
- Enterprise Adoption and Real-World Implementation
- The Risk That Controls Themselves Become a Target
- Timing and Market Conditions
- The Broader Implications for AI Infrastructure Security
- Frequently Asked Questions
Why Former SentinelOne Leaders Are Betting on Enterprise AI Security
Nicholas Warner and fellow founder Shlomi Salem, who was Vice President of research at SentinelOne, aren’t making a random pivot. Endpoint security, the domain where SentinelOne built its reputation, depends on understanding how legitimate-looking activity can mask harmful intent. That expertise translates directly to AI agents, which are fundamentally different from human employees or malware—they operate at machine speed, make decisions in milliseconds, and can touch dozens of systems in a single workflow without generating obvious alarms. The founders recognized this gap not as a distant future problem but as an immediate crisis that enterprises deploying AI today are already facing with minimal visibility.
The team also includes Eran Shirazi, formerly VP of Research at Any.do, who brings additional depth in understanding how autonomous systems behave at scale. When an AI agent starts running automatically across an enterprise, traditional monitoring tools designed for human workflows—logging suspicious login patterns, detecting unusual data access—become essentially useless. An AI agent accessing 10,000 customer records in 30 seconds isn’t suspicious by traditional metrics; it’s a routine operation if the agent was tasked with customer analysis. Neo Security’s founders recognized that detection and prevention require a fundamentally different approach when the actor is an autonomous program with legitimate access.
Building a Control Layer for Enterprise AI at Scale
Neo Security’s product is positioned as a real-time control layer for enterprise AI agents and AI-enabled software, designed to address security risks from autonomous systems operating with legitimate employee permissions. Rather than trying to detect when an AI goes rogue—a nearly impossible task when legitimate and malicious behavior look identical—the product enforces policies about what AI systems can do, much like role-based access controls limit what humans can access. The critical limitation here is that this approach requires enterprises to define those policies, which means understanding not just what their AI agents should theoretically be allowed to do, but what they actually need to do in practice.
One practical challenge that enterprises will face: overly restrictive controls can cripple AI productivity, while loose ones leave vulnerabilities wide open. A policy that prevents an expense-management AI from ever accessing payroll systems sounds secure until the AI needs to cross-reference salary data to catch fraud. The company must thread a needle between security and usability, and history suggests that needle is incredibly fine. Previous endpoint security tools have struggled for years with the tension between protecting systems and allowing them to function; Neo Security will face similar pressures, and enterprises deploying the product will need experienced security teams to tune and maintain these controls continuously.
The Investor Thesis and Series A Details
Andreessen Horowitz and Bessemer Venture Partners co-led the $75 million Series A round, with support from Craft Ventures and Merlin Ventures. The size and caliber of these investors signal that the venture community sees enterprise AI security not as a niche problem but as a fundamental market requirement. A16z in particular has been actively investing in AI infrastructure and AI safety; their participation suggests confidence that Neo Security addresses a real, near-term problem that enterprises will pay to solve. Bessemer’s track record in security investments, including early backing of SentinelOne, indicates these investors understand the competitive dynamics of security businesses and the timing required to capture market share.
This funding round arrives after Neo Security closed a $25 million seed round in 2025, meaning the company has raised $100 million total with relatively little public announcement. The stealth-mode approach—building in the background with paying customers rather than raising hype—is deliberate. Early customers in this space are sophisticated enterprises running large-scale AI deployments, not startups experimenting with chatbots. By staying quiet, Neo Security could iterate with real deployments before the inevitable wave of competitors, larger security vendors, and open-source projects all rush into this market. The question is whether that head start will matter when Microsoft, CrowdStrike, Palo Alto Networks, and others eventually enter the space with billions in resources.
Enterprise Adoption and Real-World Implementation
For a Chief Information Security Officer (CISO) evaluating Neo Security, the practical value proposition centers on visibility and control before something goes wrong. Consider a financial services firm deploying an AI agent to automatically reconcile transactions: the agent might be legitimately authorized to query transaction databases, but without a control layer, it could theoretically be prompted to export entire transaction histories or modify settlement records. Neo Security’s approach would let the CISO define that this agent can *read* transaction data but never *modify* it, that it can query transactions from the past 90 days but not older records, and that any access to customer PII needs a secondary approval. These policies translate abstract security requirements into enforceable technical constraints.
The tradeoff is operational overhead. Implementing such granular policies requires security teams to model AI workflows in advance, anticipate what data an agent legitimately needs, and then monitor and adjust as actual deployments diverge from planning. For a large enterprise with hundreds of AI agents in development, this could mean months of policy definition work before the first agent goes into production. For a startup with three people and one AI agent, Neo Security might feel like overkill. The market opportunity is clear—enterprises are deploying AI at scale and most have no adequate governance framework—but adoption speed will depend on how much easier Neo Security makes this work compared to building internal controls or using generic security tools.
The Risk That Controls Themselves Become a Target
As enterprises rely on control systems to gate what AI can do, a darker possibility emerges: what happens when attackers focus on compromising the control layer itself rather than the AI agent? If an attacker can modify the policies that Neo Security enforces, they could theoretically disable all restrictions and grant an AI unlimited access. This is the classic security principle that adding more components adds more attack surface—a control layer between AI and data is valuable only if it’s harder to compromise than the protections it provides. The warning here is subtle but critical: Neo Security’s product will need to be extraordinarily secure, because it becomes the highest-value target in an enterprise’s AI infrastructure.
Another limitation that enterprises should consider: Neo Security’s product is a technical control, not a governance solution. If a company’s problem is that it doesn’t know which AI agents it’s even running, or which executives approved them, or what data they’re supposed to be accessing, Neo Security won’t solve the organizational dysfunction that led to that chaos. The product assumes enterprises have already made decisions about which AI systems to deploy and why; it enforces those decisions technically but can’t make those decisions for you. This is fine for well-governed enterprises, but many of the companies rushing to deploy AI today are doing so precisely because governance is fragmented and speed is the priority.
Timing and Market Conditions
Neo Security’s emergence in July 2026 arrives at a specific moment in enterprise AI adoption. Large companies have moved beyond the experimental phase of testing large language models on safe, bounded problems. Now they’re deploying autonomous agents into production workflows—systems that make decisions without human approval on every action. This shift from “AI assistants that help humans” to “AI agents that act independently” is the trigger that makes security solutions like Neo Security suddenly urgent.
Vendors like Databricks, Hugging Face, and various enterprise AI platforms are shipping agent frameworks; enterprises building with these tools are running into the reality that giving an AI agent broad permissions creates security gaps that existing tools don’t address. The funding size also reflects market enthusiasm ahead of actual proof points. Neo Security had paying customers before the Series A closed, but we don’t know if those customers have deployed the product at scale or whether they’ve renewed licenses. The next 18-24 months will be critical for demonstrating whether enterprises actually prioritize this control layer once their AI projects encounter the friction of policy definition and enforcement, or whether they’d rather move fast and deal with governance later.
The Broader Implications for AI Infrastructure Security
Neo Security’s funding and positioning suggest that enterprise AI security is becoming a distinct market, separate from traditional application security or endpoint security. This specialization makes sense—AI agents behave differently from code, train on different threat models, and require different monitoring and control mechanisms. However, it also fragments the security landscape further. An enterprise deploying AI now faces a decision about whether to buy specialized AI security tools like Neo Security, push their existing security vendors to solve this problem, or build internally.
Each path has costs and risks, and no path is clearly superior yet. The involvement of former SentinelOne executives is also a reminder that security businesses often iterate through multiple generations of founders and ideas. SentinelOne disrupted traditional endpoint protection; some of its founding team is now attempting the same disruption in AI security. Whether Neo Security succeeds will depend not just on its founders’ previous track record but on whether the problem it’s solving is as fundamental as endpoint security was, and whether the company can execute well enough to capture market share before larger competitors arrive with their own solutions.
Frequently Asked Questions
How much did Neo Security raise, and from whom?
The company raised $75 million in Series A funding from Andreessen Horowitz and Bessemer Venture Partners, with support from Craft Ventures and Merlin Ventures. Total funding to date is $100 million, including a $25 million seed round in 2025.
What is Neo Security’s product designed to do?
The product is a real-time control layer for enterprise AI agents, designed to monitor and restrict what autonomous AI systems can access and do, even when they have legitimate permissions within enterprise networks.
Who are the founders and executives behind Neo Security?
CEO Nicholas Warner previously served as President and COO of SentinelOne. Chief Product Officer Shlomi Salem was formerly Vice President of Research at SentinelOne. VP Research Eran Shirazi previously led research at Any.do.
When did Neo Security emerge from stealth?
The company emerged from stealth in July 2026, with the Series A announcement on July 20, 2026.
Why is AI agent security suddenly urgent for enterprises?
Enterprises are moving from experimental AI use cases to deploying autonomous agents in production workflows. Unlike AI assistants that support human decisions, autonomous agents make independent decisions with access to sensitive data, creating new security risks that traditional security tools don’t address.
What’s the difference between Neo Security’s approach and traditional access control?
Traditional access control limits what users can access. Neo Security focuses on limiting what autonomous AI agents can do with those same permissions, enforcing real-time policies about data access, modification, and scope of autonomous actions.