Israel’s Cyber Fifty — The Companies That Make a Cyber Superpower, Ranked

The liability is equally real. Cellebrite sold to Russia, Belarus, Hong Kong, China, Bangladesh and Myanmar before cutting them off; after Amnesty showed Serbian police using its tools to unlock a journalist’s phone and plant spyware, it suspended Serbian customers — but called similar Citizen Lab findings on Jordan and Kenya “speculation”, and publishes no vetting criteria. Dream was co-founded by NSO’s co-founder, Shalev Hulio, whose indictment a Barcelona court ordered in 2025; an investigation found 12 staff who had worked at NSO or other spyware firms; its “sovereign” systems come from a company with about 85% of its staff in Israel.

There is a deeper caution in the library. The October 7 failure was “not the result of a single glaring failure but rather the accumulation of several problems”; among them, a non-commissioned officer in 8200 warned in July 2023 that a Hamas exercise closely followed the attack plan, and her superiors dismissed the analysis as “aspirational” (Wyss, CTC Sentinel 2024). A system that selects talent brilliantly does not thereby guarantee it listens to it.

The decision for a buyer state: legitimacy is part of the product — demand systematic, published end-use vetting, and when buying “sovereign” capability insist on source-code access, local hosting and staff, and clean exit clauses. “Sovereignty rented from a foreign vendor is still dependency” (Dream Security dossier).

10. What the cautionary cases teach

Hunters and Deep Instinct fell out of the fifty, but their failure modes run through the list:

1. Right architecture, no distribution. Hunters bet early and correctly on the data-lake SIEM, with Snowflake, Databricks, Okta, Cisco and Microsoft’s M12 as investors. It had no telemetry of its own; the endpoint platforms turned SIEM into a bundle. No financing since January 2022; a newsroom silent since March 2025.

2. A technical lead without a platform. Deep Instinct applied deep learning to raw malware bytes — an idea traceable to Bar-Ilan research funded by a state consortium (David & Netanyahu 2015) — and won an HP OEM deal that brought “volume, not ownership”. The lead lasted about five years.

3. Hiring to the round. Deep Instinct’s 400-person plan, Hunters’ planned doubling, Cybereason, Coro.

4. A workflow moat eroded by AI. Snyk and Checkmarx lost standing as coding agents and frontier models absorbed code scanning; Pentera is now outgrown by AI-native Horizon3.ai.

5. Forecasting valuations in public. Deep Instinct’s “several unicorns within nine months”, Cymulate’s “unicorn within two years”.

6. Forgetting that security vendors are targets. Aqua’s open-source Trivy scanner was hijacked in March 2026, and the stolen credentials were reused to poison Checkmarx’s GitHub Actions.

The system’s resilience is the counterpoint. Cybereason’s founders came back as 7AI; Hunters’ CTO now lists its seed investor as his affiliation; Deep Instinct’s own dossier counts the engineers who moved on to the next wave as its lasting contribution. The decision: judge an ecosystem by how cheaply it recovers from failure — how fast talent is re-absorbed — not by the size of the rounds it celebrates.

The Fifty, tier by tier

What follows is the reference layer: one entry per company, in ranked order — what it is, why it sits where it does, and the one thing to learn. Status is as of 22 September 2026. Each entry points to its full ~1,000-word dossier in the companies/ folder.

Tier I — The Titans (1–6)

The six companies that define what the Israeli system can produce at full scale: two of the three largest exits in the country’s history, its oldest champion, its most valuable independent, and its largest listed challenger.

01. Wiz — Cloud security (CNAPP) · acquired by Google for $32B in cash, closed March 2026. An agentless security graph across AWS, Azure, GCP and Oracle that surfaces the handful of “toxic combinations” among 100,000 findings. The largest Israeli exit ever and the fastest-growing security company on record: $1M to $100M ARR in about 18 months, over $1B in 2025. It refused Google’s $23B and got $32B eight months later. Lesson: a repeat team plus zero deployment friction compounds. → 01 - Wiz.md

02. CyberArk — Identity security / privileged access · acquired by Palo Alto Networks for ~$25B, closed February 2026; now “Idira”. The vault for an organisation’s all-powerful accounts, stretched by acquisition from humans (the core product) to machines (Venafi) to AI agents. Ranked second for two decades of category ownership and a painful licence-to-SaaS shift that left 88% of $1.44B ARR as subscription. Lesson: own one control point — least privilege — and extend it across every identity type. → 02 - CyberArk.md

03. Check Point Software Technologies — Network security platform · public (Nasdaq: CHKP), ~$13.7B. It invented the commercial stateful firewall in 1993 and became Israel’s founder academy. Ranked on history, ~40% operating margins and $4.2B of cash, not growth: Q2 2026 revenue rose 1% and product revenue fell 14%. Under CEO Nadav Zafrir it is buying AI-security teams (Lakera, Cyata, Deepchecks). Lesson: profitability becomes a trap when rivals reinvest in adjacent categories. → 03 - Check Point Software Technologies.md

04. Cyera — AI-native data security, now a data-and-identity layer for AI agents · independent, $12B (June 2026). Agentless AI classification that tells a company what sensitive data it holds and who — or which agent — can reach it; the $1B Oasis purchase adds non-human identity. The most valuable independent Israeli cyber company — at a reported ~80x ARR, and loss-making. Lesson: re-run a proven architecture (Wiz’s agentless graph) on a new asset. → 04 - Cyera.md

05. SentinelOne — Autonomous endpoint, XDR and AI SIEM · public (NYSE: S), ~$8.3B. One lightweight agent on its own data lake; half of its $1.22B ARR now comes from beyond the endpoint. The largest independent Israeli-founded pure play still growing above 20% — but still at a −31% GAAP operating margin and below its ~$9B IPO valuation. Lesson: outsiders can win; founders turned down by Israeli VCs went straight to the US, with the CEO as sole salesman. → 05 - SentinelOne.md

06. Armis — Asset visibility and exposure management · acquired by ServiceNow for $7.75B in cash, closed April 2026. Agentless discovery of every connected device, from printers to infusion pumps, checked against a knowledge base of some 7 billion devices. Sold at about 23x ARR while growing over 50% at $340M+, six years after Insight’s $1.1B buyout. Lesson: find the pain before the product — its founders cold-called CISOs before choosing one. → 06 - Armis.md

Tier II — The Category Leaders (7–18)

Companies that own a category, or created one, but have not reached Titan scale — because they are still private, too narrow, already absorbed, or carrying governance or growth questions.

07. Cato Networks — SASE · independent, $4.8B; ARR $415M+, growing ~42%. Networking and security written as one cloud stack on its own global backbone; a Gartner SASE Leader three years running, and Shlomo Kramer’s third company. Below the Titans because it is private, smaller than Palo Alto’s $1.5B SASE line, and its IPO keeps slipping. It is tripling its Prague R&D centre. Lesson: replace a budget (the WAN) rather than add a line item. → 07 - Cato Networks.md

08. Island — Enterprise browser · independent, $4.8B; ~$200M revenue growing ~100% (founder-stated). A Chromium browser with a policy engine at the last mile — copy, paste, download and AI-prompt control without VDI or VPN; eight of the ten largest banks use it. It created the category; it sits here because its figures are unaudited and Palo Alto’s Prisma Browser is bundling the idea. Lesson: take over a default application instead of adding an agent. → 08 - Island.md

09. Varonis — Data security · public (Nasdaq: VRNS), ~$5.6B; Proofpoint reported in acquisition talks. Twenty years of permissions metadata with automated remediation — an unglamorous file-access problem made urgent by AI copilots. SaaS is now 95% of revenue after a transition that drove the stock to a $19.70 low. Growth in the high teens keeps it mid-tier. Lesson: boring structural problems make durable franchises. → 09 - Varonis.md

10. Claroty — Cyber-physical systems security · independent, ~$3B; ARR over $200M. Built inside the Team8 foundry and sold through its investors Siemens, Rockwell and Schneider to plant engineers who distrust IT vendors. With Armis gone to ServiceNow, it is the leading independent in OT, IoT and medical-device security — still pre-profit, and well short of the ~$500M revenue a listing now needs. Lesson: strategic investors can be your sales force. → 10 - Claroty.md

11. Axonius — Cyber asset and exposure management · independent, $2.6B; ARR over $200M. A neutral, agentless layer that correlates 1,400+ data sources to answer “what do we actually have?”, with a FedRAMP-authorised federal arm serving 90+ agencies. Close to cash-neutral; it raised $200M flat by choice. Held back because it is single-category and its likely buyers are its competitors. Lesson: pick the unsexy, universal problem. → 11 - Axonius.md

12. Upwind — Runtime-first cloud security · independent, ~$3.8B (September 2026). It sees what actually executes inside cloud workloads and is the leading independent alternative to Google-owned Wiz. Built by the Spot.io team after its sale to NetApp. The fastest-rising valuation of 2026 — $900M to $3.8B in 21 months — with no ARR disclosed, hence #12. Lesson: when the category leader is acquired, the independent number two gets a window. → 12 - Upwind.md

13. Imperva — Web application, API and data security · acquired by Thales for $3.6B, December 2023. Two decades of web-application and database-firewall IP, and two exits (Thoma Bravo at $2.1B, then Thales). Its founders went on to Trusteer, Incapsula and Cato, and Israel remains Thales’ largest cyber hub. Ranked on franchise and founder yield rather than independence. Lesson: a first company that trains founders is worth more to an ecosystem than its own exit. → 13 - Imperva.md

14. Cellebrite — Digital forensics for law enforcement · public (Nasdaq: CLBT), ~$2.4–2.8B. The global default for extracting court-admissible evidence from seized phones: ~86% gross margin, 117% net retention, ARR of $508M. Documented misuse by customers in Serbia, Jordan and Kenya, a 2026 guidance cut and three CEOs in about twenty months keep it mid-table. Lesson: when you sell to governments, legitimacy is part of the product. → 14 - Cellebrite.md

15. BioCatch — Behavioural biometrics for banks · Visa agreed to acquire for $2.4B (August 2026, pending). It reads how a customer types, swipes and holds the phone to catch scams in which password, device and customer are all genuine — 19 billion sessions a month across 350+ banks. EBITDA breakeven at $100M ARR, then Permira, then Visa. One vertical and one signal keep it at #15. Lesson: sell to the budget that feels the loss. → 15 - BioCatch.md

16. Forter — E-commerce fraud prevention · independent, $3B (2021, stale). A one-second approve-or-decline decision drawn from a cross-merchant identity graph, backed by a chargeback guarantee: it sells an outcome, not a score. Mid-table because it sits beside core cybersecurity and has disclosed neither a new price nor revenue since 2021. Lesson: underwrite your own accuracy — a guarantee turns a risk tool into a revenue product. → 16 - Forter.md

17. Snyk — Developer-first application security · independent; last priced at $7.4B, now marked ~$3.7B; interim CEO. It created developer security with a free tier and automatic fix pull-requests, reaching $325M+ ARR. Growth slowed to about 12%, its Israeli team shrank to about 90, and coding agents now threaten the workflow it owned — its Evo agent layer is 60% of new deals. Lesson: a workflow moat lasts only as long as the workflow. → 17 - Snyk.md

18. Radware — DDoS and application protection · public (Nasdaq: RDWR), ~$1.24B. Carrier-grade DDoS protection in appliance and cloud form; profitable, with $400M+ in liquid assets, founder-run for 29 years, and cloud ARR past $100M growing 22%. Long-run growth of 3–4% a year and related-party governance hold it at #18. Lesson: durability is a strategy — but venture bets incubated inside a slow public company (SkyHawk) rarely survive. → 18 - Radware.md

Tier III — The Scale-ups and Specialists (19–32)

Unicorns and near-unicorns of two vintages: the 2021 cohort still growing into peak prices, and the 2025–26 cohort raising on the AI thesis.

19. Torq — AI SOC · independent, $1.2B (January 2026). AI agents triage and respond across whatever tools a customer already runs, on a no-code automation engine that was ready when LLMs arrived. Repeat founders (Luminate, sold to Symantec) with Check Point engineering roots. Last disclosed ARR was $24M+ (2024) against a $100M target. Lesson: build the plumbing before the magic. → 19 - Torq.md

20. Orca Security — Agentless cloud security · independent, $1.8B (2021). It invented and patented SideScanning — full cloud coverage in minutes — and then lost the scale race to Wiz, which used the same idea with faster enterprise selling. About 350 staff now carry a five-year-old mark. Lesson: being first is not enough; weigh go-to-market ability as heavily as technical originality. → 20 - Orca Security.md

21. Silverfort — Identity protection · independent, ~$1B (January 2024). It enforces MFA inside the authentication stream itself, reaching legacy apps, command-line admin tools and service accounts that other products cannot; it claims 10 billion authentications analysed a day. No ARR disclosed since “tens of millions”. Lesson: build at the point of enforcement, and solve the ugly legacy problem first. → 21 - Silverfort.md

22. Transmit Security — Customer identity, passwordless and fraud · independent, $2.2B (2021). Login, identity orchestration and fraud scoring for banks such as Citi, JPMorgan and HSBC — Mickey Boodaei’s third company after Imperva and Trusteer. No new capital in five years, no revenue disclosed, and a $543M preference stack any exit must clear. Lesson: a serial founder’s most durable asset is the buyer relationship. → 22 - Transmit Security.md

23. Aqua Security — Container and cloud-native security · independent, above $1B (flat since 2021). It pioneered container runtime security in 2015; its open-source Trivy scanner is the developer funnel — and in March 2026 was hijacked to steal CI/CD secrets. Its founders handed over to a sales-led CEO in 2025. Lesson: open source is distribution and attack surface at once; release engineering is a security function. → 23 - Aqua Security.md

24. Checkmarx — Application security testing · owned by Hellman & Friedman ($1.15B, 2020). A static-analysis engine embedded in 60% of the Fortune 100, founded out of the Mamram and Matzov units, with its CTO still in place after twenty years. A sale process seeking $2.5B+ has produced no deal, and AI is compressing classic scanning. Lesson: private equity is a real exit path for a capital-efficient enterprise vendor. → 24 - Checkmarx.md

25. Pentera — Automated security validation · independent, ~$1B (flat since 2022). It runs real exploits safely in production networks — a one-day proof of value that ends in domain admin — and was first in its category to $100M ARR. AI-native Horizon3.ai now outgrows it, and two 2026 layoff rounds cut about 20% of staff. Lesson: automate the job you did yourself — then add ML talent before AI redefines it. → 25 - Pentera.md

26. Salt Security — API security · independent, $1.4B (2022). It made API security a budget line, using behavioural baselines of API traffic to catch logic abuse, and now pitches “agentic security”. It is the last large standalone player after Noname and Traceable were absorbed, on about $75M of reported revenue. Lesson: ask early whether your category can survive on its own. → 26 - Salt Security.md

27. Dream Security — AI-native national cyber defence · independent, $3B (June 2026). It sells state-actor detection and “sovereign AI” to governments, with ARR above $100M within three years, largely through a former Austrian chancellor’s access. NSO lineage, a founder facing a Spanish indictment order, and a very small customer base cap it at #27. Lesson: in government markets distribution is the moat — and investors must price the tail. → 27 - Dream Security.md

28. Glow — AI-native endpoint security · independent, $1.2B (out of stealth July 2026). AI agents make application allow-listing — a category that failed because humans could not maintain the lists — workable, deciding which coding agents and MCP servers may run on a laptop. $180M raised; no revenue disclosed; ranked on team and thesis. Lesson: revive a “right idea, wrong decade” category once technology removes its constraint. → 28 - Glow.md

29. Zenity — Security and governance for AI agents · independent; $125M Series C (August 2026). It finds, governs and blocks AI agents across Copilot, ChatGPT, Claude and Cursor; it began in 2021 on low-code governance, and the market grew into its category. Revenue triples yearly from an undisclosed base, and Microsoft is both channel and competitor. Lesson: choose a wedge that sits just ahead of a bigger trend. → 29 - Zenity.md

30. Zafran — Threat exposure management and mitigation · independent; over $140M raised. It asks whether a “critical” vulnerability is running, reachable, already blocked or actively exploited — and then mitigates through controls the customer already owns. Founded by two decorated 8200 majors and a Unit 81 technologist; ARR has tripled, but is still “in the millions”. Lesson: reframe the question and the backlog shrinks. → 30 - Zafran.md

31. Noma Security — AI and AI-agent security · independent; $132M raised. One platform to find agents — typically 10 to 100 times more than teams expect — map their tools and MCP connections, red-team them and govern them at runtime. ARR up 1,300% from an undisclosed base, with Zenity as a direct Israeli rival. Lesson: build the angel round out of the CISOs you want as customers. → 31 - Noma Security.md

32. Oligo Security — Runtime application and AI security · independent; ~$140M raised. A kernel-level eBPF sensor that sees whether the vulnerable function inside an open-source library ever executes — and blocks the exploit if it does. Its research disclosures (ShadowRay, 0.0.0.0-Day) are its marketing. About 100 staff in a runtime market dominated by platforms. Lesson: go one level deeper than the incumbents look. → 32 - Oligo Security.md

Tier IV — The AI-Era Vanguard and the Niche Leaders (33–50)

Two kinds of company share the bottom tier: very young firms ranked on team, thesis and early momentum, and long-lived specialists that dominate a narrow market. Both are where the next Titans — and the next cautionary cases — will come from.

33. 7AI — Agentic SOC · independent; $130M Series A at ~$700M. AI agents triage and investigate alerts on top of a customer’s existing tools, sold channel-first through DXC. Founded by two of Cybereason’s co-founders; every metric is self-reported and the company is under two years old. Lesson: talent from a failed company comes back — pick the channel on day one. → 33 - 7AI.md

34. Irregular (formerly Pattern Labs) — Frontier AI security lab · independent; ~$450M (2025). It tests unreleased models from OpenAI, Anthropic and Google DeepMind for offensive cyber capability, and its CEO co-authored RAND’s SL1–SL5 model-weight security levels. About 25 staff; in 2026 its test environments reached the real internet. Lesson: sell to the people who create the risk — and run offensive-AI test ranges as critical infrastructure. → 34 - Irregular.md

35. Alice (formerly ActiveFence) — AI safety and security · independent; $140M at ~$700–800M (August 2026). Eight years of labelled real-world abuse data turned into red-teaming and guardrails for eight of the ten leading AI labs; ARR approaching $100M. A mixed, services-heavy business, priced accordingly. Lesson: an old dataset can be the ticket into a new market. → 35 - Alice.md

36. Sygnia — Incident response and MDR · owned by Temasek (~$250M, 2018). The ex-8200 team boards call mid-breach, turning crises into retainers and managed detection. A Team8 foundry company sold at about 58 times its capital within three years; two CEO changes in 2025. Lesson: services can yield a venture-scale exit if you sell to the top of the market and exit early. → 36 - Sygnia.md

37. XM Cyber — Exposure management and attack paths · owned by Schwarz Group (~$700M, 2021); CrowdStrike buying its IP. It models an estate as a graph and finds the “choke points” where attack paths converge — an offensive planner’s view from a former Mossad director. It will now license back its own technology. Lesson: make the attacker’s mental model the product. → 37 - XM Cyber.md

38. Zero Networks — Automated microsegmentation · independent; over $100M raised. It makes a proven but consultant-heavy control deployable in weeks: agentless rules generated automatically, MFA at the network layer. Revenue doubled, net retention above 120%, 99% of customers retained — from a small base. Lesson: look for proven ideas that failed on deployment. → 38 - Zero Networks.md

39. Apiiro — Application security posture management · independent; ~$135M raised. Deep code analysis maps software architecture and flags only material, risky changes; it is now moving into governing AI coding agents. It walked away from a $500–600M Palo Alto deal in 2022 and grew ARR 104% in 2025. Lesson: repeat founders carry a method, not a product. → 39 - Apiiro.md

40. Cycode — ASPM and software supply-chain security · independent; ~$81M raised. Its own scanners plus a code-to-runtime context graph; a Leader in Gartner’s first Software Supply Chain Security Magic Quadrant (2026). Capital-efficient, but without a priced round since 2021. Lesson: enter through the gap incumbents ignore — build pipelines and leaked secrets in 2019. → 40 - Cycode.md

41. Cymulate — Breach and attack simulation · independent; ~$500M (2022). Easy SaaS attack simulation across 1,000+ customers, now generating detection rules and fixes as well as tests. Born inside a penetration-testing services firm; it missed its own public unicorn deadline. Lesson: productise the services work you keep repeating — and never forecast your valuation. → 41 - Cymulate.md

42. Astrix Security — Non-human identity and AI-agent security · acquired by Cisco for ~$400M (June 2026). It named “non-human identity” early and extended one identity graph from SaaS tokens to AI agents, selling within five years of founding on about $85M raised. Lesson: a short category name buyers repeat can do more than features. → 42 - Astrix Security.md

43. Koi Security — Software supply-chain and “agentic endpoint” security · acquired by Palo Alto Networks, closed April 2026. It inventories and controls what users install for themselves — browser extensions, IDE plugins, packages, MCP servers — which endpoint tools mostly ignore. Founded in 2024 and sold about twenty months later: reported at ~$400M, though Palo Alto’s 10-K shows $231M plus $61M of replacement equity, roughly twice its Series A price. Lesson: prove the attack before you sell the defence — its fake VS Code theme reached 300+ organisations in a week. → 43 - Koi Security.md

44. Sweet Security — Cloud and AI runtime security · independent; ~$120M raised. An eBPF sensor plus cloud logs that now blocks rogue AI-agent behaviour, founded by the IDF’s former CISO after he failed to find a good enough cloud detection tool for the army’s move to the public cloud. ARR up sixfold from an undisclosed base. Lesson: start from a buyer problem you owned yourself. → 44 - Sweet Security.md

45. Legit Security — ASPM and AI-generated-code security · independent; $77M raised. It maps the “software factory”, ranks findings with its own scanners, and governs coding agents such as Claude Code and Cursor. Founded by Checkmarx veterans; no round since 2023. Lesson: aggregate, but own something — and use offensive research as marketing. → 45 - Legit Security.md

46. KELA — Cybercrime threat intelligence · independent (Vector Capital minority stake). Fifteen years of underground data plus human analysts; bootstrapped for nine years and profitable since inception; bookings up 101% last year; its research fed the TeamPCP arrests. Lesson: proprietary data compounds, and public-interest work is a sales channel. → 46 - KELA.md

47. Cynet — All-in-one XDR with bundled MDR · independent. One agent, one console and a human SOC in the box for companies that have no SOC; MSPs brought 64% of its Q4 2025 deals, on about $78M of disclosed funding. Lesson: build for the buyer the leaders underserve, and bundle the service into the software. → 47 - Cynet.md

48. Guardz — SMB security and MDR sold through MSPs · independent; $84M raised. One multi-tenant console that lets an outsourced IT provider secure dozens of small clients — identity, email, endpoint, cloud data — with AI triage and a human MDR team behind it. It owns the glue and rents the engines: SentinelOne (also an investor) for endpoint, Check Point for email. ARR grew 500% and then 300%, from an undisclosed base. Lesson: sell to the intermediary, not the end customer — and note that exits create founders; its own came from IntSights and Argus. → 48 - Guardz.md

49. Coro — Unified SMB security platform · independent; ~$275M raised. One agent and many modules for lean-IT mid-market firms, sold 100% through partners since 2025. Hypergrowth was followed by a reset and a channel veteran replacing the founder as CEO. Lesson: heavy capital in a small-ticket market becomes a burn problem the moment growth slows. → 49 - Coro.md

50. Waterfall Security Solutions — OT unidirectional gateways · independent. Hardware that lets data leave a power plant or rail network while nothing can come back in — an approach NERC CIP, US nuclear rules and France’s ANSSI reward or require. About 100 staff after nineteen years. Lesson: make the regulator your sales team, and sell certainty where failure is physical. → 50 - Waterfall Security Solutions.md

The cautionary cases

C1. Hunters — SOC platform / data-lake SIEM · no financing since January 2022. Right about architecture, backed by Snowflake and Databricks, beaten by platforms that own telemetry. → C1 - Hunters.md

C2. Deep Instinct — Deep-learning malware prevention · headcount down from ~400 to ~180. A genuine technical edge that never became a scaled business. → C2 - Deep Instinct.md

What this means for us

The Czech Republic starts from a better position than its size suggests — and with a clearer diagnosis of its gap than most. NÚKIB’s 2026 strategy lists real strengths: an advanced national system, internationally respected experts, a working community across state, business and academia, and the observation that “some of the most widely used commercial ICT security tools are developed by Czech companies or have domestic origins”. It is equally frank about the weaknesses: staffing and funding “insufficient in both the public and private sectors”, rigid public-sector pay, a still-reactive state, and an EU-wide shortage of up to 300,000 experts (NÚKIB 2026). The country has publicly attributed attacks to Russia (May 2024) and China (May 2025). It has the talent and the threat. What it lacks is the machine that turns the one into companies.

What to copy

• Founder academies, deliberately. Treat anchor companies and foreign R&D centres as schools for the next founders, and make leaving easy. Microsoft’s and Symantec’s Israeli centres produced Wiz, Apiiro, Zero Networks, Torq and Cycode.

• Specialist seed capital with a CISO network. One domain-expert fund, Cyberstarts, backed Wiz, Cyera, Island and Upwind early. Pair such funds with global VCs in “split-seed” rounds — twelve in Israel in 2024 (YL Ventures 2024).

• A foundry wired to buyers. Team8 started from a thesis and a network of corporate security chiefs and produced Claroty and Sygnia. Our equivalent is a foundry tied to the operators regulated under the new Cyber Security Act.

• The state as the demanding first customer. Axonius built a federal arm around FedRAMP; Waterfall made regulators its sales team. In Europe, NIS2 and the new Cyber Security Act are our FedRAMP: rules that turn security into a budget line.

• An alumni engine for our own units. 8200’s formal alumni association (14,000+ members) and its accelerator turn service into a network (Rousseau 2017). The Czech Cyber and Information Warfare Command and NÚKIB can build the same.

• A wide intake. Israel’s Magshimim trains gifted coders from underprivileged areas (IISS 2021), and many of the fifty came from outside 8200 altogether — SentinelOne, Checkmarx, Cymulate, Forter, Irregular.

What not to copy

• Spyware and inconsistent end-use vetting. The NSO lineage is a permanent liability, as Dream’s diligence file shows. Cellebrite’s uneven response to Amnesty and Citizen Lab findings shows how an ethics process that looks arbitrary gets priced in.

• Peak-cycle mega-rounds and hiring to the round. Transmit, Snyk, Cybereason and Deep Instinct are the evidence.

• Selling anchors with nothing negotiated. CyberArk cut about 500 jobs within days of closing. Snyk’s Israeli team fell to about 90.

• Faith in technical superiority over judgement. Before October 7, an 8200 non-commissioned officer’s warning was dismissed as “aspirational” (Wyss, CTC Sentinel 2024).

• A narrow talent base. Only 23% of Israel’s AI professionals are women (OECD 2025). NÚKIB’s commitment to equal opportunity for women and non-graduates is the better starting point.

• Rented sovereignty. “Sovereign” systems built by a foreign vendor are still dependency — and so is European hosting of US-owned IP, the arrangement under which CrowdStrike agreed to buy XM Cyber’s patents from its German owner.

Five moves for the next twelve months

1. Stand up a cyber foundry with ten critical-infrastructure buyers. NÚKIB convenes energy, banking, health and transport operators as design partners and first customers. The foundry picks theses from their unsolved problems and recruits founding teams against them, Team8-style. Target: three companies incorporated within the year.

2. Seed a specialist cyber fund with the state as anchor LP. The fund needs a domain-expert general partner with a CISO network, and a mandate to co-invest alongside tier-one international funds at seed. It should be judged on follow-on rounds raised, not on deployment speed.

3. Recruit Israel’s overflow, on spin-out-friendly terms. Cato is already tripling its Prague centre, and Israel’s talent ceiling pushes its scale-ups abroad. Make Prague and Brno the default second R&D site for Tier II–IV companies, with fast work permits and R&D tax treatment. Pair that with employee-option taxation at exit and no punitive non-competes, so today’s hires become tomorrow’s founders.

4. Make the state a demanding first customer, with sovereignty conditions. Create a procurement track for young domestic vendors of the “secure technological alternatives” NÚKIB calls for. Fund NIS2-driven SMB security programmes delivered through managed service providers — the channel Guardz and Coro were built for, and a market the enterprise-focused ecosystem neglects. Foreign “sovereign” suppliers should have to provide source-code access, local hosting and staff, published end-use vetting and clean exit clauses.

5. Build a national AI-security gym and disclosure pipeline. The library finds no Israeli public equivalent of DARPA’s AIxCC or the Five Eyes’ CAGE challenges, and flags this as a gap a mid-sized EU state could fill (Zhang et al 2026; Standen et al 2021). A Masaryk University researcher already co-authored NATO’s reference architecture for autonomous cyber-defence agents (Kott et al 2019). A cyber range where defensive AI agents are trained, tested and certified, plus a coordinated-disclosure pipeline through the national CERT, would put Czech teams in the evaluation-and-assurance niche. That is where categories are still being named.

For founders, the fifty compress into five habits: name the category, remove the deployment objection, make the American seller a co-founder, map the likely buyers from the first round, and never forecast your valuation. For investors: pay up for repeat teams, ask for ARR before paying for momentum, model every exit at normal multiples, and treat private equity as an exit, not a failure.

Close

Israel’s fifty are not a list of lucky start-ups. They are the visible output of a machine: it selects talent early, trains it on state-scale problems, passes it through anchor companies and specialist funds, and recycles every exit into the next founding team. In 2025–26 the machine delivered its best year and exposed its weakness in the same stroke: the champions it builds are sold, and strategic control goes with them.

A mid-sized European state does not need its own Wiz to benefit. It needs the conditions in which the second-time founder stays, the foreign R&D centre becomes a founder school, and the state is a demanding first buyer rather than a grant office. Those conditions are copyable, cheap next to what they return, and slow to build. The option value lies in starting now: every year of delay is another cohort of engineers trained in Prague and Brno whose first company gets built somewhere else.