Cybertech's Sixteen Principles — What Israel Teaches the World About Building Cyber Power

7. Keep people circulating — army, lab, company, and back

• The principle in one line. Make the border between state, academia and industry porous, trusted and free of intellectual-property claims, and keep the traffic dense enough to carry tacit knowledge.

• The Israeli evidence. The least-celebrated mechanism is reserve duty. The IDF computer school’s roughly 400 reservists contribute about 20,000 days a year, equal to 100 work-years, and teach at least 97% of the advanced classes. Industry experts share tacit knowledge there that they withhold in paid settings, and the IDF claims no “property rights or patents on technologies developed under its sponsorship” (Breznitz 2002). Churn is designed in. About a quarter of 8200 turns over each year, reservists return for up to three weeks a year into their forties, and the 15,000-member alumni association exists for business and hiring rather than commemoration (Cordey 2019). INSS calls the result “double-feeding”: the same people circulate between army, industry and academia (Antebi 2021). Anchor firms such as Check Point and foreign R&D centres such as Microsoft’s and Symantec’s became founder academies (principle 12). Density matters too. 120 firms sit on about 8 sq km of central Tel Aviv, where every Israeli cyber firm with more than 100 staff is based (Hatuka and Carmel 2021).

• Why it works. Circulation moves the knowledge that documents cannot carry, and it does so in both directions: the state learns what industry has built, and industry learns what the state actually needs. A no-IP rule removes the friction that would otherwise stop the flow. Density makes each exchange cheap.

• Where it fails or is contested. You cannot decree a cluster. Beersheba got a prime-ministerial declaration, the national CERT, a university centre, a technology park and tax discounts, and still went from one start-up founded in 2013 to eight in 2020, against 337 in Tel Aviv; its status is “not advancing, and perhaps is even eroding” (Getz, Buchnik and Zatcovetsky 2024). Only 19% of the relevant career military staff said they would move south, and 66% preferred Tel Aviv (Gal et al 2017). Circulation also has a club problem: knowledge stays inside a reservist “club membership” (Breznitz 2002), and recruitment skews to the Tel Aviv area and elite schools (Cordey 2019). And in war the circuit reverses: about 10% of high-tech and management staff were called up after October 2023 (Taub 2025).

• How to transfer it. Czechia’s strategy already names enhanced civilian-military cooperation and “making use of non-state capabilities to address crises” (NÚKIB 2026). Estonia has already shown an EU-and-NATO version of the mechanism: the Defence League’s Cyber Defence Unit, a volunteer body of public- and private-sector professionals given a statutory mandate in 2014 (Osula, CCDCOE 2015). The missing pieces for us are the mechanics: a standing reserve of civilian experts with clearances handled once, a no-IP rule on state-funded training and tooling, and the discipline to concentrate the traffic in Prague and Brno rather than spreading it thin across the regions.

• The first move. Stand up a cyber reserve corps of about 200 named industry experts contracted for ten to fifteen days a year to teach on state courses, staff crisis rotas and play in sector exercises, with security clearance processed once and centrally. Publish a rule that the state claims no IP on training materials or tools produced under its sponsorship.

8. Widen the funnel before it narrows you

• The principle in one line. Decide who belongs in cyber at fourteen, not at twenty-four, because belonging — not ability — is what determines who enters.

• The Israeli evidence. In a survey of 912 Israeli ninth-graders, 36% of Jewish boys, 18% of Arab/Palestinian boys and 6% of girls in both sectors wanted a technology job at thirty. The strongest predictor was identity: students who felt similar to programmers were more than nine times more likely to aspire to such a job, while liking mathematics explained nothing (Budge, Charles, Feniger and Pinson 2023). The adult numbers follow. About 94% of high-tech workers are non-Haredi Jews and about a third are women (Taub 2025), and only 23% of AI professionals are women (OECD 2025). Breznitz called the exclusion of Arab and ultra-Orthodox citizens “the vast waste of human resource from a national point of view” as early as 2002. Israel knows it: Magshimim targets the periphery, and the 2025 strategy commits to widening entry and to hiring wounded veterans into junior roles (INCD 2025). The counter-examples show what is possible. About 70% of staff in the 1991–98 incubator wave were recent immigrants (Trajtenberg, NBER 2000), and Zafran’s chief executive arrived as an Iranian-born teenage refugee and rose to major in 8200.

• Why it works — when it does. Widening the intake is the cheapest capacity increase available, because the constraint is people, and because the marginal recruit from an excluded group is competing against nobody. It also reduces concentration risk in the economy: Israeli high-tech workers pay about 6.3 times more income tax than other workers and provided about 36% of all wage-based income tax in 2021 (Taub 2025). A narrow base carries a whole state.

• Where it fails or is contested. This is the principle Israel has least delivered on. The gender gap is larger in Hebrew-language than Arabic-language schools, and the cultural association of computing with Jewish military service reinforces it (Budge et al 2023). Structural exclusion from the military route cannot be fixed by an after-school course. And the consequences are now visible as brain drain, more company registrations in the US and net emigration of skilled workers (Taub 2025).

• How to transfer it. NÚKIB has already written the commitment: inclusive training, and explicit attention to “equality of employment opportunities and conditions for women, experts without a university degree and graduates of non-technical disciplines” (NÚKIB 2026). What the Israeli research adds is the measurement point. Do not measure diversity at the hiring stage, where the funnel has already closed. Measure it at fourteen, and treat “feeling similar to a programmer” as the policy target.

• The first move. Run an annual survey of technology aspiration among fourteen- and fifteen-year-olds, disaggregated by gender and region, and publish it as one of NÚKIB’s indicators. Tie the funding of new education programmes to moving that number, and open every state-run cyber course to applicants without a university degree.

III · The Capital and the Market: the company machine

Israel’s cyber industry was not the natural result of having clever engineers. Venture capital emerges only on top of pre-existing research and human capital, and even then it has to be engineered (Avnimelech, Kenney and Teubal 2004). Israel engineered it in the early 1990s with about $100M and a good contract, then rebuilt the layer twice: first as specialist seed funds and foundries, and now, as private fundraising thins, as public patient capital again. The four principles in this cluster are the most directly copyable in the report, and the cheapest.

9. Share the upside, never insure the downside

• The principle in one line. Public money should take a minority position, bring in a foreign partner, and let private investors buy the state out cheaply if they win.

• The Israeli evidence. Yozma (1992–93) was a $100M government fund. It put $8M into each of ten private funds, never more than 40% of any fund, and required each fund to bring in an established foreign financial institution, because the skills and networks “did not exist in Israel”. Private partners held a five-year option to buy out the state’s stake at cost plus 5–7% interest. The designers’ logic was that “successes should be highly profitable, while failure would be subsidized only to the extent that the government investments were lost” (Avnimelech, Kenney and Teubal, BRIE 2004). Venture investment rose from $5M in 1990 to $3.3B in 2000. Yozma-descended funds held about 55% of the pool and had the highest survival rates through the 2001–03 crash. Its predecessor, Inbal (1992), insured the downside and failed. The same logic ran through the incubators, where the state recovered its money only if a venture succeeded (Noël 2020), and through the 1984 R&D Law’s neutral, demand-driven grants (Trajtenberg 2000). Israel is now re-running the play. Yozma 2.0 (2024) pays institutional investors extra yield to enter venture capital (OECD 2025), and a NIS 250M fund-of-funds inside the Yozma Fund targets deep tech (IIA 2025).

• Why it works. Asymmetric terms make the state a cheap, temporary partner rather than a shareholder to be managed. The mandatory foreign partner imports judgement and networks alongside the money, and officials sitting on fund boards learn what they are subsidising. Neutrality — funding any qualifying project rather than a chosen sector — reinforces emerging comparative advantage instead of guessing at it.

• Where it fails or is contested. Yozma worked because decades of defence R&D and university training had already produced the deal flow; sequencing matters. The agency that invented it then lost most of its freedom: its budget allocation by 2005 was “barely above a third” of the 2000 level (Breznitz and Ornston 2012). Fund clocks are a structural problem in their own right — with ten-year lives, general partners abandon good but slow projects, at an efficiency loss that “can be in excess of 25%” (Kandel, Leshchinskii and Yuklea 2005). And the model has not produced durable domestic scale: new Israeli tech companies fell from a peak of 1,404 in 2014 to about 500 in 2024 (IIA 2025).

• How to transfer it. The Czech gap is not the absence of money but the absence of terms. NÚKIB’s strategy records that funding conditions and EU rules “are unclear and continue to grow in complexity”, that procedural barriers stop organisations drawing available funds, and that NÚKIB will build “a professional structure for providers of state aid for research and development whose programmes focus on cyber security” (NÚKIB 2026). A Yozma-style instrument is a single contract, not a new agency.

• The first move. Design one fund-of-funds instrument on Yozma terms: state money capped at 40% of each fund, a mandatory tier-one international co-investor in every fund, a five-year buy-out option at cost plus a small interest rate, and a state observer on each board whose job is to report what is being learned. Anchor three or four specialist funds with it, and publish the terms before the call.

10. Put domain experts behind the first cheque

• The principle in one line. The first investor should know the buyer’s problem better than the founder does, and should bring the buyer with them.

• The Israeli evidence. A handful of domain-expert seed investors shaped a generation. Cyberstarts wrote the first cheque into Wiz and backed Cyera, Island, Upwind, Glow, Zafran and Legit; Wiz alone returned it around $1.5B. YL Ventures seeded Axonius with $4M in 2017 and sold the stake for $270M in 2021. Team8, founded in 2014 by 8200 veterans with corporate backers from Microsoft to Temasek, works as a foundry. It starts with a thesis, then recruits founders, and opens its network of corporate security chiefs from day one. It produced Claroty and Sygnia, which raised about $4.3M and sold to Temasek for a reported $250M within three years. The model has become market structure. In 2024 Israeli cyber had a record 50 seed rounds worth $400M, tier-one global funds joined 22 seed rounds (against 10 a year earlier), and 12 “split-seed” rounds were co-led by an Israeli specialist and a global fund (YL Ventures 2025).

• Why it works. Domain investors can judge whether a problem has a budget line before it has a product, and their networks turn the first customer conversation into design input. A foundry compresses the riskiest year of a company’s life: the thesis is validated by buyers before a founder is hired. Split seeds combine local judgement with the global follow-on capital that the local market cannot supply.

• Where it fails or is contested. The same layer of specialist capital also financed the other Israeli cyber industry: the Atlantic Council counts 14 US investors in spyware vendors, 12 of them in Israel (Atlantic Council 2024). Specialist funds also concentrate risk in fashionable categories — new firms crowd into cyber and fintech even as formation collapses (IIA 2025) — and the Series A and B layer thinned in 2024 while seed boomed. Peak-cycle prices did real damage: Transmit Security’s $543M Series A and Snyk’s 2022 valuation both became exit constraints. And even a well-backed thesis can lose: Hunters was right about the data-lake SIEM and beaten by platforms that owned the telemetry.

• How to transfer it. For a country our size, the scarce asset is not capital but a general partner with a security-chief network and the standing to convene buyers. That person can be recruited, and the state can make the job attractive by pairing the Yozma instrument with privileged access to the operators regulated under the Cyber Security Act. Report 1 sets out the foundry model in detail; the principle here is narrower — never let public money be the only money, and never let it be the expert money.

• The first move. Assemble a register of 30 chief information security officers from regulated operators who agree to act as design partners and, under published conflict-of-interest rules, as angel investors. Make state co-investment conditional on a round being co-led by a domain specialist and an international fund — the split-seed rule, written into the term sheet.

11. Borrow the reach a small country lacks

• The principle in one line. A small state cannot manufacture proximity to the world’s customers, capital, research and intelligence, so it should import them deliberately and on terms.

• The Israeli evidence. The cluster’s first economic job was to overcome distance from the market. Venture capital, US-based management, acquisitions as “a shortcut to the market” and early Nasdaq listings gave Israeli firms the reach their geography denied them (de Fontenay and Carmel, Stanford SIEPR 2001). Yozma imported foreign partners for their know-how. About 300 multinational R&D centres (Noël 2020), 47 of them in cyber (ICDK 2020), became founder schools. The company template is explicit: Island builds 95% of its product in Israel with a US executive team, and Salt Security’s lesson is to “hire the US seller as a co-founder, not as an early employee”. Reach is also state business. Israel “lacks the global intelligence reach” of the largest powers and compensates through its alliance with the US (IISS 2021). The US–Israel Cybersecurity Cooperation Enhancement Act (2016) created a DHS grant line for joint ventures, with at least 50% private cost share, merit review and governance by the binational research foundations. The INCD works with more than 90 CERT partners and calls cyber defence “a global ‘team sport’” (INCD 2021).

• Why it works. Imported reach is cheaper than building a domestic market that does not exist. Multinational centres bring frontier problems to local engineers and then lose those engineers to start-ups. Alliances buy early warning that no small state can collect alone. And capacity building abroad turns technical excellence into diplomatic standing (Tabansky 2016).

• Where it fails or is contested. Borrowed reach means borrowed control. Four of the five largest acquirers of Israeli cyber firms are in the San Francisco Bay Area (Hatuka and Carmel 2021). The dual-headquarters model leaves the headquarters, sales and much of the value in California, and R&D can drain too: Snyk’s Israeli team fell from several hundred to about 90 while the company passed 1,500 staff. Alliance dependence cuts both ways — the same United States that amplifies Israeli capability blacklisted NSO and stopped the L3Harris deal.

• How to transfer it. Our borrowed reach is institutional: NATO, the EU and the single market. NÚKIB’s strategy is explicit that for resource-constrained states “the solution ... lies in intensive cooperation with reliable partners”, names cyber attachés and the NATO Integrated Cyber Defence Centre, and singles out Czechia’s “multi-layered relationship with Israel” (NÚKIB 2026). Cato Networks is already tripling its Prague R&D centre. The task is to convert relationships into instruments with money attached.

• The first move. Create a binational co-funded industrial R&D instrument with Israel on the DHS–BIRD model: joint ventures only, at least 50% private cost share, merit review, unclassified projects, and a sunset clause. Pair it with a landing programme that makes Prague and Brno the default second R&D site for scale-ups, on the condition that local teams are allowed to spin out.

12. Recycle every exit

• The principle in one line. Treat an acquisition as the start of the next three companies, and design tax, liquidity and R&D rules so that the people and the money stay.

• The Israeli evidence. Recycling is the most consistent finding across the fifty dossiers. Microsoft’s ~$320M purchase of Adallom (2015) produced both Wiz and Armis. Symantec’s purchase of Luminate produced Torq, NetApp’s purchase of Spot.io produced Upwind, and Aorato’s founders went on from Microsoft to build Apiiro. Failure recycles too: Cybereason’s collapse produced 7AI, and IntSights and Argus alumni built Guardz. Check Point plays the “Fairchild” role, its alumni founding or seeding Palo Alto Networks, Imperva, SentinelOne and Illusive (Hatuka and Carmel 2021). In 2024, 15 new Israeli cyber start-ups were founded by entrepreneurs who had already sold a company, and five Israeli start-ups bought other Israeli start-ups (YL Ventures 2025). Wiz’s sale alone is estimated to have produced about NIS 10B in Israeli tax revenue, roughly $1.5B for one seed fund and about 1,000 newly liquid engineers.

• Why it works. An exit returns three things at once — capital, experienced operators and buyer relationships — and it returns them into a network dense enough to reuse them within months. The acquirer’s local R&D centre then becomes the next founder school, which is why the Imperva dossier’s lesson for states is to negotiate R&D continuity at exit rather than to mourn the deal.

• Where it fails or is contested. The exit is also the leak. Between 1995 and 2005, 89% of Israeli tech firms followed one of two paths, venture-only or venture-then-trade-sale, and only about 6% touched public markets (Hashai and Menuhin 2006). “The products created by the Israeli digital industry are in fact the companies themselves” (Noël 2020). Decision rights go with the sale: CyberArk cut about 500 jobs within days of closing, and Israel no longer has an independent anchor in either cloud or identity. The OECD names the structural cause as “a comparatively low base of long-term capital” (OECD 2025), and the Taub Center now records contraction in start-up formation, skilled emigration and a rise in US company registrations.

• How to transfer it. Czechia will not have a $32B exit soon, but it already has acquisitions, foreign R&D centres and senior engineers with option packages. The policy surface is small and specific: taxation of employee options at exit rather than at grant, enforceable limits on non-competes, room for secondary sales so staff can take liquidity without forcing an early sale (the Island and Claroty pattern), and commitments on local R&D negotiated when a state-supported company is bought.

• The first move. Write an “exit covenant” into every state-backed instrument: any company that has taken public research or venture money commits, on acquisition, to maintain its local R&D headcount for a defined period and to report it. Pair it with a matched pre-seed facility open only to founders and senior engineers who start a new company within twenty-four months of an exit, and publish an annual register of where R&D and decision rights actually sit.

IV · The Guardrails and the Frontier: legitimate and ahead

The first twelve principles explain how Israel built capability and companies. The last four explain what keeps them ahead of adversaries and able to operate in the world. Two are about the frontier: design against a real attacker, and publish what you learn. Two are about the guardrails. This is the cluster Israel executed least evenly, and the evidence for it is Israel’s own worst years.

13. Specify security against a named adversary

• The principle in one line. For every asset, first ask which actor it must withstand, then build and test to that tier. Publish a national threat of reference and revise it every year.

• The Israeli evidence. Resolution 3611 made an annual “national threat of reference” one of the bureau’s nineteen founding tasks, and Israeli cyber policy grew from national-security threats rather than cybercrime (Benoliel 2015). The habit reaches the frontier. Pattern Labs, now Irregular, co-wrote with RAND the model-weight security framework used by leading AI labs. It identifies about 38 attack vectors and sets five levels, from SL1 (amateurs) to SL5 (the most capable nation-states) (Nevo et al, RAND 2024). Offence taught defence. Stuxnet’s roots were “not in the IT domain but in nuclear counter-proliferation”, and in industrial systems “the worst vulnerabilities are not bugs, they are features” (Langner 2013). The defensive record follows. Iran has been “forced to focus mainly on soft targets” (Anderson and Sadjadpour, Carnegie 2018). An Iranian attack on the grid in 2015–16 hit honeypot decoys, and wartime DDoS peaking at a million attempted logons a second in 2023 had no significant impact (Freilich, INSS 2024). In June 2025 Israel “emerged largely unscathed in the cyber domain” (Sharma, MP-IDSA 2025). Firms sell the same logic: Waterfall sells “certainty to buyers facing physical consequences”, and XM Cyber made “the attacker’s mental model the product”.

• Why it works. A named adversary makes security investment proportionate, because not everything needs SL5. It gives buyers and vendors a specification that can be tested, and it lets offensive knowledge inform defence.

• Where it fails or is contested. A named adversary can harden into a “conception”. Hamas showed no cyber spike before 7 October (Google TAG and Mandiant 2024), and low-end vectors were missed. Palestinian Islamic Jihad intercepted unencrypted IDF drone feeds for two years and hacked road cameras for rocket targeting. In June 2025 Iranian actors used weakly secured Israeli security cameras to adjust missile fire (Freilich 2024; Sharma 2025). Ransomware locked every system at Hillel Yaffe hospital in October 2021 and bed occupancy fell from 83% to 64% (Abbou et al 2024); a threat model built around states has to cover that too. A perimeter defined by named entities also misses consumer IoT: 1,355 hijacked smart sprinklers could empty a water tower in an hour (Nassi et al, BGU 2018).

• How to transfer it. NÚKIB already names its adversaries. It calls Russia “the biggest direct and long-term threat” and also names China, North Korea and Iran, and it records Iranian activity against Czech water-management infrastructure. Czechia made its first public attributions in May 2024 and May 2025 (NÚKIB 2026). What remains is to turn those names into specifications. The comparative literature adds a budget rule: rank critical infrastructure by importance, so a smaller state protects its most important assets first (Tvaronavičienė et al 2020).

• The first move. NÚKIB issues an annual national threat of reference, classified in full with a public summary. Starting with water utilities, each designated operator assigns an adversary tier to its most critical systems and has them tested against that tier. Cameras, remote-access paths and consumer-side devices count as in scope.

14. Publish the attack with the fix

• The principle in one line. Fund deep, open research in a few elite laboratories, and make every attack they publish arrive with a coordinated disclosure and a mitigation that can be deployed.

• The Israeli evidence. The base is decades of theory. Shamir co-invented RSA (Rivest, Shamir and Adleman 1978). Biham and Shamir’s differential cryptanalysis rediscovered in public an attack that IBM and the NSA had kept secret for eighteen years (Biham and Shamir 1993). Goldreich and Wigderson co-proved that every NP statement has a zero-knowledge proof (Goldreich, Micali and Wigderson 1991). After a 2012–13 funding call, matched, excellence-based academic cyber centres produced more than 500 papers (Cohen et al, New America 2017; ICDK 2020; INCD 2021). The signature is attack plus fix. Tel Aviv University’s NXNSAttack found DNS amplification above 1,620 times. BIND, Unbound, PowerDNS and CZ.NIC’s Knot shipped fixes, and Google, Cloudflare, Amazon and Microsoft patched, mostly with the authors’ MaxFetch mitigation (Afek, Bremler-Barr and Shafir 2020). Rogue7 was disclosed to Siemens alongside publication (Biham et al 2019). The Technion’s Morris II AI worm came with a guardrail that caught it at a true-positive rate of 1.0 (Cohen, Bitton and Nassi 2024). Companies copy the habit: Claroty’s Team82, Oligo, Zenity Labs in OWASP, and Koi, whose thirty minutes on the Darcula attack produced its thesis, first product and press coverage.

• Why it works. Publishing an attack together with its fix earns trust from vendors and regulators and sets standards. It also turns a lab into a school: Daniel Genkin, whose acoustic key-extraction work with Shamir and Tromer produced a GnuPG patch, went on to co-author Spectre. Start-ups get their theses from the same work, and in the dossiers research is repeatedly “the go-to-market”.

• Where it fails or is contested. The academic base is thinning. Israel’s share of world publications fell from 0.94% to 0.67% between 2000 and 2022, only 1.3% of its papers are highly cited (last among eight peers), and international co-authorship is falling (Getz and Barzani, Neaman 2024). Industry funds only about 7% of university research, and more than 90% of commercialisation income comes from one institute (Bentur et al 2019). The capability is dual-use: people who can break ciphers are also recruits for offence. Israel also has no public equivalent of DARPA’s AI Cyber Challenge or the Five Eyes’ CAGE competitions for autonomous defence (Standen et al 2021; Zhang et al 2026).

• How to transfer it. Czech institutions are already in the loop. CZ.NIC’s Knot was among the resolvers fixed after NXNSAttack, and Masaryk University co-authored NATO’s reference architecture for autonomous cyber-defence agents (Kott et al 2019). NÚKIB coordinates national cyber research. Its own strategy admits, though, that “the practical use and deployment of research results is often problematic” (NÚKIB 2026). The international frameworks list what is missing. The Oxford model’s top stage of responsible disclosure requires legal protection for those who disclose flaws responsibly (Oxford GCSCC 2021). ENISA asks: “Is there a plan to link R&D initiatives with real economy?” (ENISA 2020).

• The first move. Run a matched-funding call on Israeli terms for two or three university cyber research centres: awards on excellence criteria, with each university matching the funding. Add a disclosure clause: every attack a centre publishes goes through coordinated disclosure via the national CERT and ships with a mitigation. Legislate a safe harbour for good-faith security research in the same session.

15. Build the guardrails with the capability

• The principle in one line. Put export control, end-use vetting and domestic oversight in place while offensive-grade skill is being built, not after, because the reputational cost falls on the whole ecosystem.

• The Israeli evidence. Resolution 3611’s work plan already called for better export procedures and oversight in 2011. What followed shows the cost of letting capability outrun governance. Eight Israeli vendors make up 43.9% of the 435 entities in the Atlantic Council’s global spyware dataset (Roberts et al 2024). Of the 74 governments known to have bought spyware or forensics tools, 56 bought from firms based in or connected to Israel. A study cited by Haaretz found that 80% of the founders of Israel’s roughly 700 cyber companies had served in IDF intelligence (Feldstein and Kot, Carnegie 2023). Candiru “reportedly recruits from the ranks of Unit 8200” and ran more than 750 sites impersonating Amnesty, the WHO and others, and Israel’s export licensing is “almost entirely opaque” (Citizen Lab 2021). Licensing followed geopolitics. Ukraine was blocked from buying Pegasus and Estonia barred from using it against Russian targets. Israel is not a formal Wassenaar member, and when rules tightened, firms moved export desks to Cyprus (Feldstein and Kot 2023). The check that worked came from outside: the US Entity List designation of November 2021 pushed NSO towards bankruptcy, while regulators at home seemed “largely unfazed” (Kotliar and Carmi 2024). At firm level, Cellebrite suspended Serbian customers but dismissed similar evidence on Jordan and Kenya, an inconsistency its dossier says investors price in.

• Why it works — where it is done. Guardrails are a licence to operate in allied markets. They protect clean firms from being tarred by others, keep the alliance that supplies reach (principle 11), and give talented people an ethical line. Magshimim teaches “cyber ethics” for this reason.

• Where it fails or is contested. Export control is also foreign policy, and secrecy limits transparency. Demand does not go away: when one supplier is sanctioned, buyers move to others, and vendors change jurisdiction and name. Candiru has traded under at least five names.

• How to transfer it. Czechia is today more buyer than producer, which makes this the cheapest moment to act. 14 of the 27 EU member states bought from NSO (Atlantic Council 2024). The European Parliament recommends import controls, an EU entity list and more transparency. The Atlantic Council recommends “Know Your Vendor” rules for government buyers, beneficial-ownership registries and published export licences. NÚKIB’s strategy already commits to regulation “balancing national security and individual rights”, with powers subject to monitoring (NÚKIB 2026).

• The first move. Attach a guardrail clause to every state instrument proposed in this report: the fund, the pilots, the R&D centres and the binational fund. The clause requires a published end-use policy for dual-use products, disclosure of beneficial ownership, and no sales of intrusion tools to unvetted end users. The state, as a buyer, adopts “Know Your Vendor” rules and publishes an annual aggregate report on intrusion-tool procurement.

16. Protect the dissenter from the machine

• The principle in one line. Automate collection and triage, but do not automate away the person who notices what the model does not expect. Warnings should be pushed to decision-makers, not left in a pool.

• The Israeli evidence. October 7 was “not the result of a single glaring failure”, but its causes included “an overestimation of Israel’s technological capabilities” and a drift from human to technical collection. Monitoring of Hamas hand-held radios reportedly stopped a year earlier, while Hamas used wired phones and fed deception through channels it knew were watched. The IDF had held the “Jericho Wall” attack plan for more than a year. A Unit 8200 NCO who warned in July 2023 that Hamas exercises matched it was dismissed as “aspirational” (Wyss, CTC Sentinel 2024). A former senior officer blames a “love affair” with cyber intelligence and AI. In his account, automated processing “with little or no human intervention” replaced pushed warnings with an “intelligence pool”, and 8200’s open-source unit, Hatzav, was dismantled in 2021 on the strength of machine translation (Bar, NIPP 2024). The chief of staff had warned in 2018 that “too much information is not a guarantee for better command and control” (Eizenkot 2018). Civilian cyber defence, by contrast, held through 2023–25. The technical literature agrees that machine-learning defences can never be the only layer, because adversarial training “should have no effect” on adversarial examples (Shamir et al 2019).

• Why it works. A record of technical superiority breeds a fixed belief about what the enemy will do, and automation removes the friction where anomalies used to surface. A protected dissenter puts that friction back on purpose. It matters most exactly when it feels least necessary, because the riskier a surprise attack looks, the more readily it is dismissed (Wyss 2024).

• Where it fails or is contested. Dissent channels decay into ritual. Israel created a “Devil’s Advocate” unit after 1973 (Rousseau 2017), and fifty years later such mechanisms had become “routinized and ritualized” (Wyss 2024). A speak-up culture in a hierarchy is “easier in theory than in practice”. Israel’s 2025 strategy now writes in countermeasures against the biases that lead analysts to underrate strategic surprise (Shabtai 2025), but a written rule is not proof that behaviour has changed.

• How to transfer it. NÚKIB plans AI-based detection and the automation of routine work so that savings can be redirected to salaries (NÚKIB 2026). That is the right direction, provided each automation decision is weighed against the human capability it removes, and the analyst who disagrees has a guaranteed route to the top.

• The first move. Adopt a written push rule for the national CERT, the National Cyber Operations Centre and the intelligence assessment staff. Any analyst may escalate a warning to the director and must receive a written reply within 72 hours, and every escalation is logged and reviewed each year. Before any analytic unit is cut or replaced by automation, a red team reviews which human capabilities will be lost.

What not to copy

Some of Israel’s advantages are side effects of conditions no democracy should create on purpose. Others are choices that Israel’s own research now treats as mistakes. The following should stay in Israel.

• The threat as the engine. Conscription-scale selection and a permanent existential threat gave Israel its workforce advantage (”Israel’s tech-heavy military and national conscription gave it an advantage”, Lewis 2016). A state cannot import them, and should not try. Czechia’s route to scale runs through coalitions, EU funds and certification.

• Spyware as an export industry. Opaque, geopolitics-driven licensing made Israel-linked firms the supplier to 56 of 74 known spyware-buying governments, invited a Cyprus workaround and imposed a reputational cost on the whole ecosystem. No one should copy it.

• Deniable tit-for-tat as deterrence. Israel-linked personas have taken Iranian petrol stations and banks offline (Google TAG and Mandiant 2024; Sharma 2025). But Israel’s reported counter-strike after the 2020 water attacks did not deter Iran, whose attacks resumed within weeks (Freilich 2024), and in the 2025 war cyber gave only an “incremental edge” (Sharma 2025). NÚKIB commits to developing offensive and multi-domain capabilities. They belong under law and allied coordination, not in signalling games.

• Powers ahead of oversight. The 2018 draft bill would have allowed equipment seizure without a court order and monitoring of all internet traffic. Israel’s legal dimension is its weakest, rated only “established” (Lewis 2016).

• The exit-only default and peak-cycle capital. Selling anchors without negotiating anything, $543M Series A rounds and hiring to the size of the round (Transmit, Snyk, Deep Instinct) are costs of the Israeli model, not features of it.

• A cluster by decree. Beersheba has a declaration, a CERT, a university and tax discounts, and saw eight start-ups founded in 2020. Buildings do not move people.

• A closed elite. A pipeline that reproduces one demographic runs out of people and concentrates national risk in a narrow, mobile workforce.

The principles as a system

The sixteen principles are not a menu. Israel’s advantage comes from how they feed each other, and its current weaknesses show where the loop leaks. The flywheel runs in six steps:

1. Ends and an architect (1, 2) create a mandate and remove the turf friction that stops everything else.

2. A trusted civil defender and a demanding state customer (3, 4) turn the threat into problems that companies can solve and data that defenders can share.

3. The talent engine (5–8) turns those problems into people who have already solved them at national scale.

4. Capital and reach (9–11) turn those people into companies with global customers.

5. Exits (12) return money, experienced operators and buyer relationships to the start of the loop, along with tax revenue that funds step one.

6. The frontier and the guardrails (13–16) keep the problems hard, the research ahead of the attacker, and the licence to operate intact.

Five pairs do most of the reinforcing:

• 4 and 13: a named-adversary specification is what makes the state a demanding customer rather than a grant office.

• 3 and 14: a trusted, no-fault CERT is also the channel for coordinated disclosure.

• 6, 7 and 12: missions, circulation and exits together produce the repeat founder, the most valuable asset in the dossiers.

• 9 and 10: Yozma’s foreign partners created the venture layer from which the specialist funds later grew.

• 11 and 15: alliances supply reach, and guardrails keep them. The US Entity List showed that an ally can also close the door.

The loop now leaks at three points. The narrow funnel (8) is starving the pipeline: the number of new companies has halved from its peak, skilled workers are emigrating and more companies are registering in the US. Exits (12) are sending decision rights abroad. And the dissent failure (16) contributed to a surprise that Israeli analysts now compare with 1973 (Bar 2024). A state copying the model should build these three points more strongly than Israel did, not less.

If a state is starting today, four principles matter most:

1. Treat talent as national infrastructure (5). It has the longest lead time: a fifteen-year-old recruited in 2027 is a founder in 2037. It is also the constraint every comparative study names.

2. Make the state the first demanding customer (4). It is the cheapest principle, because it uses rules already in force. It is also the fastest, because pilots can start this year.

3. Share the upside, never insure the downside (9). A single well-drafted contract brings in the capital and the know-how a mid-sized market lacks, and Israel is running the same play again with Yozma 2.0.

4. Build the guardrails with the capability (15). It is almost free at the start and ruinously expensive to retrofit, as NSO showed.

Principle 2 is the precondition for all four. Without market pay and the authority to arbitrate between ministries, none of them gets done.

How to know it is working. Measure capability, not commitment. The ITU index put Israel below Estonia and beside Czechia, and ENISA concedes that its own framework measures maturity rather than effectiveness (ENISA 2020). Use outcome indicators instead: the technology aspiration of fourteen-year-olds, operator-course graduates placed each year, paid pilots with young vendors, split-seed rounds, companies started by founders after an exit, disclosures shipped with fixes, and median time from disclosure to mitigation in regulated entities. Top that list with the Oxford model’s own definition of the highest stage for a national cyber marketplace: “Domestic cybersecurity products are exported to other nations and are considered superior products” (Oxford GCSCC 2021).

Close

Israel’s cyber power is often explained by its enemies, its army or its culture. The evidence points elsewhere. Across 78 nations, regional differences in cyber culture disappear once development and the scale of internet use are accounted for (Creese, Dutton and Esteve-González 2021). Israel became a cyber power before it had a strategy, then built the machinery to keep that power and turn it into an industry: a NIS 4.5M bureau, a civil defender without police powers, a six-month course designed backwards from a real mission, a $100M fund with a buy-out clause, and matched academic centres that publish the fix with the attack. Where Israel failed, the causes were choices, not fate: a narrow funnel, opaque export licensing, and faith in machines over the analyst who disagreed.

A mid-sized European state can copy the mechanisms without the threat, and it can build the guardrails that Israel added late. Czechia starts with more than it thinks: a comprehensive cyber law since 2014, a respected agency, a strategy that already names the right gaps, and Czech code and researchers inside the global disclosure loop. What it lacks is the engine: the pipeline, the terms, the first customers and the recycling that turn talent into capability and companies. Each of those is cheap next to the threat it answers, and each takes years to mature. The option value lies in starting all four first moves this year, because every year of delay is another cohort of fifteen-year-olds who never find out that the field was meant for them.