Data loss prevention is a field defined by frustration. Security teams spend months writing complex rules to govern what data can leave a network, only to watch those policies generate thousands of false positives or miss novel leaks entirely. The fundamental problem is static policy versus dynamic human behavior. Orion Security, a Tel Aviv-based startup, is betting that the answer is not better rules, but no rules at all. Its platform uses AI agents to analyze the full context of every outbound data transfer in real time, deciding autonomously whether to block it. The bet has already convinced Norwest Venture Partners and IBM to lead a $32 million Series A, bringing the company's total funding to $38 million [Calcalist, May 2026].
The wedge of autonomous context
Orion's core argument is that legacy DLP tools fail because they lack situational awareness. A policy might flag an employee emailing a spreadsheet to a personal address as a violation, but it cannot discern if that employee is a CFO sending quarterly projections to their personal accountant or a salesperson sharing a public pricing sheet with a spouse for formatting help. Orion's system, according to company statements, examines a wider set of signals: the sensitivity of the data, the user's identity and role, their apparent intent, the purpose of the transfer, and the historical patterns of both user and recipient [Calcalist, May 2026]. An AI agent synthesizes this context in milliseconds to make a block/allow decision, aiming to stop actual leaks while permitting legitimate work. This shift from policy-based to agent-based enforcement is the company's primary wedge into the entrenched DLP market, which is dominated by large vendors like Symantec, McAfee, and Forcepoint.
Traction and the investor signal
Orion moved quickly from its $6 million seed round in March 2025 to its Series A just over a year later [PRNewswire, March 2025]. The company reported ramping from zero to seven-figure annual recurring revenue within five months of launching its platform [LinkedIn - Or Daniel, 2026]. While specific customer names are not public, Orion states it serves organizations in heavily regulated finance, healthcare, and technology sectors [JPost, May 2026]. The investor roster provides strong early validation. Norwest Venture Partners, a firm with deep enterprise software experience, led the Series A. Perhaps more strategically significant is the participation of IBM, a giant in regulated-industry IT with its own extensive security portfolio. This suggests Orion is being evaluated not just as a point solution, but as a potential component or go-to-market partner for larger platforms.
The founding team brings a blend of technical depth and go-to-market experience relevant to the problem.
| Role | Name | Background |
|---|---|---|
| CEO | Nitay Milner | Based in New York. Former roles at Forbes, Cisco, and infrastructure observability company Epsagon [LinkedIn, 2026]. |
| CTO | Jonathan Kreiner | Based in Tel Aviv. Former roles at Forbes and WalkMe, with a technical background from Israeli Military Intelligence Unit 8200 [LinkedIn, 2026]. |
The technical breakdown and scale risks
The promise of agentic DLP is a dramatic reduction in administrative overhead and false positives. The technical reality is that it replaces the known complexity of policy management with the black-box complexity of model inference. Orion's agents must be trained on vast, diverse datasets of both malicious and benign data transfers to understand context accurately. A failure in training data coverage, such as a novel exfiltration method or an uncommon but legitimate business process, could lead to catastrophic false negatives or positives.
At scale, three specific challenges emerge. First is latency. Adding a real-time LLM inference step to every outbound network packet, email, or file upload is computationally expensive. Performance under peak load in a large enterprise could degrade network throughput if not architected perfectly. Second is explainability. When a traditional DLP rule blocks a transfer, an admin can point to the violated policy clause. When an AI agent blocks it, the "why" may be buried in a neural network's weights. This is a critical barrier for adoption in audit-heavy environments like finance and healthcare, where regulators demand clear reasoning for security actions. Third is the data flywheel. The system's accuracy should improve with more customer data, but early customers with unique data patterns may experience poorer performance, creating a cold-start problem for new industry verticals.
Orion's early momentum is undeniable, and its technical premise addresses a genuine pain point. The company's next twelve months will be about proving its architecture can handle the throughput of a global bank, providing the audit trails that satisfy compliance officers, and expanding its contextual understanding beyond its initial verticals. If it can navigate those scaling risks, it won't just be selling a better DLP tool, it will be redefining how enterprises think about data boundary control.
Sources
- [Calcalist, May 2026] Cyber startup Orion raises $32 million Series A backed by IBM for AI-driven data-leak prevention | https://www.calcalistech.com/ctechnews/article/byzup00kw11e
- [The Jerusalem Post, May 2026] Israeli startup ORION raises $32 million in Series A round | https://www.jpost.com/business-and-innovation/tech-and-start-ups/article-885400
- [PRNewswire, March 2025] Orion Security launches from stealth with $6M to secure enterprise data from insider threats using generative AI | https://www.prnewswire.com/il/news-releases/orion-security-launches-from-stealth-with-6m-to-secure-enterprise-data-from-insider-threats-using-generative-ai-302404662.html
- [LinkedIn, 2026] Or Daniel - ORION Security | https://www.linkedin.com/in/or-daniel-864018140/
- [LinkedIn, 2026] Nitay Milner - Orion Security | https://www.linkedin.com/in/nitay-milner/
- [LinkedIn, 2026] Jonathan Kreiner - Orion Security | https://www.linkedin.com/in/jonathan-kreiner-8b2bb81ab/