The Russian Cyber Threat and How Israel Treats It

Microsoft closes the loop from telemetry. Russian state actors “appear to have reduced their efforts to develop bespoke operations in favor of leveraging the cybercriminal ecosystem”, which Microsoft reads as a response to public exposure of their tooling — and warns that it “could make it more difficult for network defenders to attribute simple operations to sophisticated threat actors and recognize the implications of a breach” (Microsoft 2025, p. 48). Commodity malware on your network no longer implies a commodity adversary. Tkachuk adds the reverse flow: Russian intelligence services provide criminal groups with “access to advanced technologies, vulnerabilities and cyberattack tools, including those targeting mobile devices”, raising the risk of uncontrolled proliferation of what used to be state-only capability (CCDCOE 2025, p. 16). Mandiant reaches the same judgement about APT44 itself, calling it “a significant proliferation risk” that has “lowered the barrier of entry for other state and non-state actors”. And the fuel supply for all of it is the credential economy: Russia is the world’s second-largest source of infostealer infections, at 40,868 encounters behind India’s 44,197 (Microsoft 2025, p. 22). Hence Microsoft’s summary line for 2025 — “adversaries aren’t breaking in, they’re logging in”.

7. The information apparatus is industrial, and its budget is public

The psychological limb is not a side project run by trolls. It is a funded industry with a published state budget line. The European External Action Service detected 540 FIMI incidents in 2025 involving 10,500 unique channels and websites and about 43,000 observables across 19 platforms, with 29% attributed to Russia, 6% to China and 65% unattributed but showing coordination indicators with Russian or Chinese infrastructure. More than 100 countries, around 200 organisations and roughly 140 named individuals were targeted. Russian state-media funding rises to 146.3 billion roubles (about €1.56bn) in 2026, 7% up on 2025 (EEAS 2026, pp. 5, 9–11).

The architecture is deliberately covert. Of the 3,000 most recurrent channels, only 9.5% are overt state-controlled or official outlets; 90.5% are covert state-linked or state-aligned assets (p. 9). This is the structural reason why sanctioning RT and Sputnik achieved so little: the overt tenth was never where the work happened.

Two case files show the production model. Doppelgänger cloned at least 17 authentic media brands — Bild, 20minutes, Ansa, The Guardian, RBC Ukraine — across more than 50 typosquatted domains bought between May and September 2022, carrying about 80 identified pieces of disinformation in seven languages (EU DisinfoLab 2022, pp. 2, 6). Critically, it ran inside Europe: registrars GoDaddy, NameCheap and Panamanames; hosting at Webzilla in Luxembourg, BlueVPS and Glesys in Estonia and Sweden, JavaPipe in the Netherlands; cloaking and geo-redirection through Keitaro, a tracker registered in Estonia. “They targeted European citizens through impersonating European media providers, hosted their operations on European servers and covered their traces using European software” (p. 4).

Storm-1516, documented by France’s VIGINUM in a TLP:CLEAR technical report, ran 77 information operations between August 2023 and March 2025: 35 against Ukraine and its leadership, 42 against Western personalities, events or elections, and 20 specifically denigrating candidates or attacking ballot integrity. The production line includes video and voice deepfakes, forged invoices, fabricated press articles and government documents, and recruited amateur actors in more than half of the operations, with narrators chosen to match the story — an Arabic speaker for “Ukraine recruits ISIS fighters”, surzhyk for a faked Zelensky phone call (VIGINUM 2025, pp. 3–10). The distribution chain is the real intelligence product: burner or paid accounts seed, foreign paid media launder the content into “news”, and a known network of Western pro-Russian influencers amplifies it in French, German, English, Finnish and Dutch. Seventeen of the outlets used had already been used by Prigozhin’s Project Lakhta — the tradecraft is inherited, not invented. And the state link is documented rather than assumed: John Mark Dougan, the Prigozhin ecosystem, Valery Korovin’s Centre for Geopolitical Expertise (sanctioned by the US Treasury on 31 December 2024 for directing and subsidising the creation of deepfakes) and Yury Khoroshenky, a likely officer of GRU Unit 29155.

The AI turn is measured rather than asserted. Twenty-seven per cent of 2025 EEAS incidents involved AI-related techniques, up from 41 cases in 2024 to 147 — a 259% rise (EEAS 2026, p. 13). Voice cloning and synthetic video are routine, and Portal Kombat is suspected of LLM grooming: flooding the web with low-quality multilingual content to poison model training data and retrieval. But the EEAS is honest about quality: “Threat actors prioritise quantity over quality, resulting in limited overall impact as organic engagement remains low… Storm-1516 and Overload use AI-generated videos easily identifiable as inauthentic by average viewers.” The aim is “not precision but presence”. That is what Goldstein and colleagues predicted in 2023: language models attack the three constraints on influence operations — resources, content quality and, most importantly, detectability, because copypasta is what usually gets a campaign caught (CSET, OpenAI and Stanford 2023, pp. 4, 13).

II · How it actually reaches Israel

With the adversary described, the question becomes empirical. Where does this machine point, and what arrives in Israel?

8. The targeting data: the world’s most-attacked country, and Russia is not the one attacking it

The numbers were given in the opening, but they deserve to be set out as a single scannable picture, because the asymmetry is the finding.

Israel’s total exposure · 603 observed nation-state events in Microsoft’s notification data, the highest in the Middle East and Africa and second worldwide only to the United States at 623 · 3.5% of all attacks Microsoft observes globally, third behind the US at 24.8% and the UK at 5.6% · in Microsoft’s own summary, “the primary geographical targets of nation-state activity this year were in Israel, the United States, and the United Arab Emirates” (Microsoft 2025, pp. 8, 43–44).

Who is producing it · Israel is 64% of all Iranian nation-state targeting · Israel does not appear in the top ten regions targeted by Russian state actors · Russian state targeting by sector runs government 25%, research and academia 13%, think tanks and NGOs 13%, IT 10%, energy 5%, defence industry 3%, and Ukraine alone absorbs 25% of all Russian cyber operations (Microsoft 2025, pp. 46, 48).

This is consistent with everything else in the library. In the six months before 7 October 2023, Iran accounted for about 80% of all government-backed phishing aimed at users in Israel (Google TAG and Mandiant 2024, p. 4). Chuck Freilich’s INSS assessment finds Israeli defences blocked most Iranian attempts on critical infrastructure, that Iranian hackers in 2015–16 “believed that they had succeeded” against the electric grid when they had in fact reached honeypot decoys, and that in the first six days of the October 2023 war DDoS against Israel reached one million attempted logons per second with no significant impact after three months (INSS 2024, pp. 5, 30). Carnegie’s earlier judgement still holds: “Given the sophistication of Israel’s cyber defense, Tehran has been forced to focus mainly on soft targets” (Anderson and Sadjadpour 2018, p. 48).

Israel’s threat picture is an Iranian threat picture with a Hezbollah and Hamas annex. It is not a Russian one, at the level of state operations. The rest of this section is about the three routes by which Russia reaches Israel anyway, none of which appear in a nation-state targeting chart.

9. Route one: proxies, and the redirection after 7 October

Israel sits fifth in the NoName057(16) targeting table at 4.50% of attack days, above Czechia and Poland, and it is the only non-European country in that band (Insikt Group 2025, p. 12). That alone would be unremarkable — a pro-Russian group attacking a country that is not pro-Russian. What makes it significant is the timing evidence.

Tkachuk’s account should be read twice: “following the 7 October 2023 terrorist attack in Israel, dozens of new pseudo-hacktivist groups were launched to attack Israeli infrastructure. During this period, Ukraine saw its lowest number of cyber incidents in years” (CCDCOE 2025, p. 18). Set beside her statement that there are no independent hacktivists in Russia and that the services run a turnkey methodology for deploying them by geopolitical requirement, this is not volunteer enthusiasm. It is capacity allocation — a resource Moscow controls, moved from one theatre to another within days of an event Moscow did not cause, and visible in the Ukrainian incident statistics as a trough.

Three implications follow for any state that is not Israel. First, the pseudo-hacktivist layer is a shared, reallocatable pool, so your quiet period may be someone else’s crisis rather than evidence that your defences improved. Second, the layer is monetising: by late 2023 some of these state-controlled groups had begun “monetising their cyberattack services” (p. 16), blurring the line with the criminal layer still further. Third, it does not only do DDoS. NÚKIB records a growing trend of Russian-speaking hacktivists attacking weakly secured operational-technology systems, especially in water management, filming their activity and posting it to Telegram — “there have been dozens of them, especially in Western countries, and NÚKIB registered one attack in the Czech Republic last year” (NÚKIB 2025, p. 50). That is the CyberArmyofRussia_Reborn pattern Mandiant documented in early 2024, when CARR posted videos claiming manipulation of human-machine interfaces at Polish and US water utilities and a French hydroelectric plant, and a local US official later confirmed a malfunction that overflowed a tank at one claimed victim (Mandiant 2024, p. 16).

A deniable, centrally tasked, low-skill layer that can be pointed at Israeli infrastructure one week and Czech water utilities the next is the operational reality of the Russian threat to countries outside the front line. It produces very little damage and a great deal of signal, which is exactly what it is for.

10. Route two: capability transfer to Iran

The second route is indirect in a more dangerous way, because it improves the adversary Israel actually has.

RAND’s 2023 assessment records that in exchange for Iranian drones, Russia has offered Tehran “an unprecedented level of military and technical support” including multi-role aircraft, air defence, intelligence, surveillance and reconnaissance, and cyber capabilities (RAND 2023, p. 1). The indicator list is more consequential than the headline: among the markers of expanding cooperation is “Russian provision of digital surveillance and censorship systems, reversing prior prohibition on sharing offensive cyber capabilities“ (p. 7). Moscow had a rule about not giving Iran offensive cyber tools. The rule went.

Transfers beyond cyber make the pattern legible. After Iran asked for help suppressing the Mahsa Amini protests, Russia supplied communication-surveillance capabilities, eavesdropping devices, advanced photography equipment and lie detectors, alongside a judicial-cooperation memorandum (p. 9). In August 2022 Russia launched an Iranian remote-sensing satellite, and IRGC affiliates publicly celebrated the improved ability “to monitor potential U.S. and Israeli military targets” (p. 8).

RAND’s restraint is itself part of the finding: “it is still too soon to declare that the Russian-Iranian relationship has undergone a fundamental transformation.” It remains a partnership of convenience with mistrust and competing interests — and the largest source of friction is named: “One significant fault line relates to Israel; Russia has historically cultivated a good working relationship with Israel, whereas Iranian rhetoric and national security policy characterizes Israel as one of the country’s primary enemies” (p. 19). Moscow has even “sought to distance itself from Iranian antagonism of Israel”.

That is the whole strategic geometry in two sentences. Russia arms and equips Israel’s principal cyber adversary, while carefully preserving its relationship with Israel — because the relationship buys something the transfers cannot. What it buys is the subject of point 16.

11. Route three: the Hebrew-language campaign, running since April 2023

The third route is the one that does not appear in any cyber telemetry, and it is the most thoroughly documented Russian operation against Israel in the library.

Daniel Rakov’s JISS study is built on leaked internal documents of Russia’s Social Design Agency (SDA) and Structura, whose authenticity a US Department of Justice release of September 2024 later confirmed. It describes a Hebrew-language arm of Doppelgänger running since April 2023, codenamed “Normal Israel”, steered from Sergei Kiriyenko’s directorate in the presidential administration, with Putin “probably at least vaguely aware of it” (JISS 2025, pp. 7–8, 11).

The objectives are not ambiguous. Deepen Israeli polarisation in order to get Israel to withhold military aid to Ukraine; distance Israel from the United States; move Jerusalem toward Moscow; and swing American Jewry against the Democratic Party ahead of November 2024 (p. 7). The first fake, on 18 May 2023, was an article attributed to the well-known columnist Amnon Abramovich on a clone of the news portal N12, written in poor Hebrew.

What makes the file worth a European reader’s attention is that the plan was never confined to the digital domain. SDA proposed founding a political party of Russian-speaking immigrants, hoping for three or four of the Knesset’s 120 seats as a balancing faction; costed an “Israeli office” at roughly $1.2 million; proposed spraying provocative graffiti in Arab neighbourhoods while posing as Jewish right-wing extremists; and proposed distributing leaflets in synagogues and churches, with a separate messaging campaign aimed at Israeli Arabs designed to stir violence against Jews (pp. 8, 31). Penetration of the local information space used real Israeli intermediaries: sponsored “articles” appeared in Walla, The Jerusalem Post and Channel 13, commissioned through an Israeli media figure of Russian origin who declined to name his clients, and a Russian-speaking Israeli blogger’s YouTube videos were amplified by Doppelgänger (p. 32).

The scale claimed internally is industrial: 34 million comments and 40,000 “content units” plus thousands of videos and memes in January–April 2024 alone, produced by ten specialised teams — from media monitoring and research to “exclusive reporters in Israel”, translation and distribution — running a “Comments Machine” designed to fake authentic engagement signals (p. 36). After 7 October the campaign went global and turned antisemitic: Doppelgänger’s European arms amplified content including an AI-generated video of Jews in Auschwitz apologising to Palestinians for the IDF’s actions, alongside Russia’s promotion of the claim that the Holocaust was a genocide of “the Soviet People” generally rather than of Jews specifically (pp. 21, 24).

Why Israel, and why this method? Because the environment is permissive in ways Europe’s no longer is. Russia can combine provocations with media activity “in light of a large number of Russian speakers, a large number of flights to Russia, and the absence of published restrictions on the activity of intelligence agents. The latter enjoy diplomatic immunity in Israel, unlike in Western countries, where there have been large-scale expulsions of Russian spies disguised as diplomats over the past five years” (p. 50). Israel’s Russian-speaking population is usually put at around a million; the library contains no census, but it contains something more useful — the adversary’s own market sizing, which was three or four Knesset seats and a $1.2m office. Israel protested to Moscow in July 2023. Nothing changed.

12. The dog that did not bark, and why it matters

Now the negative findings, because a threat assessment that only lists what happened is an advertisement.

There is no documented Russian state intrusion campaign against Israeli critical infrastructure in this library — no Israeli Industroyer, no Israeli NotPetya, no Sandworm operation against an Israeli grid. Israel does not appear in Microsoft’s Russian top ten. It is not named in Tkachuk’s list of Russian APT tasking, which runs Gamaredon (FSB) against Ukrainian government, Sandworm (GRU) against energy, Turla (FSB) against defence enterprises. It is not in the EEAS top-targets list for FIMI incidents: Ukraine 112 · France 107 · Moldova 94 · Germany 71 · USA 51 · UK 36 · Armenia 18 · Poland 17 (EEAS 2026, p. 10). And Rakov’s own effect assessment is sober: direct political influence in Israel is “quite limited”, the campaign runs “on a low-intensity scale”, and SDA’s claim of half a million Israeli followers is doubtful (pp. 8, 35, 50).

Three conclusions follow, and they pull in different directions, which is why each has to be stated.

First, the restraint is real and it is a policy. Given Mandiant’s assessment that APT44’s reach is already global — North America, Europe, the Middle East, Central Asia, Latin America — the absence is a decision, not a limit.

Second, restraint is reversible in a way that resilience is not. Rakov warns that “the relatively limited scope of Doppelganger’s activity in Israel might change rapidly based on the campaign operators’ assessment of the situation”. A capability withheld for diplomatic reasons returns the moment the diplomacy changes, and the defender’s lead time is zero, because nothing had to be built.

Third, and most useful to a European reader: Israel is the closest thing available to a natural experiment on what a Russian information campaign does when the technical campaign is switched off. Israel gets the psychological limb alone. Czechia, Poland and the Baltics get both. The answer the control group gives is that the psychological half arrives regardless, costs almost nothing, and is the half nobody in Jerusalem is mandated to fight.

III · How Israel treats it

Israel’s response to Russia has four layers, and only three of them are visible in cyber policy documents. They are worth taking in order of how well they work.

13. The technical layer: a system built for Iran that happens to stop proxies

Israel’s national cyber defence is set out in Report 2 and specified as a system in Report 3, so the summary here is compressed. The architecture is three layers with responsibility escalating by severity: organisations own robustness, the national CERT and sector centres own resilience, the security bodies own national defence (INCD 2017). CERT-IL in Beersheba runs a 119 hotline open to any citizen, handled around 9,000 incidents and issued about 300 alerts in a single pre-war year, and works with more than 90 international partners; CyberNet links company security chiefs in a trusted national network (INCD 2021). Critical infrastructure is regulated at the level of named entities rather than whole sectors, a lineage running from Resolution B/84 in 2002. The February 2025 strategy adds “zero significant damage to critical infrastructure”, a National SOC combining the sectoral SOCs, and “Cyber Dome”, an AI-driven national active-defence concept (Shabtai, BESA 2025).

Against the Russian-aligned proxy layer this is close to ideal, and it is worth being precise about why. NoName057(16)-class operations are volumetric and application-layer denial of service against public-facing government, transport and finance targets. Israel has a national scrubbing posture, a sectoral SOC structure, a CERT with a public hotline, and a domestic industry that sells exactly this — Radware’s DefensePro appliances and cloud scrubbing service, whose own 2026 threat reporting counted web DDoS attacks up more than 110%, sit in banks, carriers and governments worldwide. The proxy campaign against Israel after 7 October produced noise, Telegram claims and overstated damage reports, and on the available record no significant effect.

The honest reading is therefore slightly deflationary: Israel is well defended against the Russian technical vector because it is extremely well defended against the Iranian one, not because it built anything for Russia. That is the correct design when 64% of your nation-state pressure comes from one adversary — but it means the defence is adaptive to volume, not to a Russian change of policy.

14. The intelligence layer, and what 7 October did to it

The second layer is the one that would have to notice a change of Russian policy, and it is the one with a documented failure mode.

The October 2023 surprise was not a cyber failure; it was a warning failure with a technological cause. Michel Wyss records “an overestimation of Israel’s technological capabilities” among the reasons, a shift of collection toward technical means as HUMINT coverage of Hamas’s intentions declined, and a Unit 8200 NCO’s July 2023 warning that Hamas exercises matched the captured “Jericho Wall” plan being dismissed as “aspirational” (CTC Sentinel 2024, pp. 4–6). Shmuel Bar is more specific: the intelligence community’s “love affair” with cyber intelligence and AI, automated processing “with little or no human intervention”, the dismantling of Unit 8200’s OSINT unit Hatzav in 2021 on the strength of machine translation, the disbanding of document exploitation, and the replacement of pushed analyst warning with an “intelligence pool” that analysts had to pull from — and did not pull from when the contents contradicted the prevailing conception (NIPP 2024, p. 5).

Every one of those failure modes is more dangerous against an information-confrontation adversary than against a kinetic one. A Doppelgänger campaign in Hebrew produces no SIGINT signature, no malware sample and no network indicator. It produces text, published by intermediaries who are legally resident Israelis and laundered through legitimate outlets. Detecting it requires exactly the capabilities Israel had been decommissioning: open-source collection, linguistic analysis, cultural intelligence and an analyst willing to push an unwelcome conclusion upward. The 2025 strategy responds with compensatory mechanisms against the cognitive biases that underestimate strategic surprise (Shabtai 2025) — the right lesson, drawn from the wrong war, and applied to the enemy Israel expects.

15. The private layer: an industry as a de facto national sensor network

The third layer is not a policy at all. It is an externality of the ecosystem the first three reports in this series described, and it does more against Russia than any Israeli government body does.

Check Point’s threat research runs on ThreatCloud across more than 100,000 customer organisations, and its prevention-first research output is among the most-cited early-warning sources in the industry. KELA, founded by Unit 8200 veterans in 2009 and profitable from inception, sells visibility into precisely the Russian-speaking criminal economy that Brantly and Mason quantified: dark-web forums, marketplaces, Telegram channels and infostealer logs, with 192 billion intelligence items collected and 34.1 billion compromised credentials indexed in 2026 alone. Its 2026 State of Cybercrime report counted 2.86 billion credentials stolen in 2025 and 7,549 ransomware victims, up 45%; its research fed the TeamPCP arrests through the Australian Federal Police and the FBI; and in October 2025 it launched a National Cyber Resilience Suite for governments, national CERTs and police. Sygnia, the Team8-founded incident-response firm now owned by Singapore’s Temasek, publishes attribution-grade investigations — Bybit, Velvet Ant, Fire Ant — that function as public goods. Alice, formerly ActiveFence, built an eight-year corpus of labelled real-world abuse content across languages by tracking how influence and abuse actors operate at platform scale, and now sells red-teaming and runtime guardrails to eight of the ten leading AI labs.

Two things follow. The first is a genuine national advantage: a mid-sized state with a domestic threat-intelligence industry gets a sensor network it does not have to fund, and KELA sells national CERTs and police the same tools it sells to banks. The second is a dependency. The layer of Israel’s Russian threat picture that works best is privately owned, commercially motivated and sells to everybody. Recorded Future belongs to Mastercard and Cybersixgill to Bitsight; Israeli threat intelligence is consolidating into global platforms on the same trajectory. A national capability that exists only as a by-product of an export industry is one the state does not control.

16. The diplomatic calculation: Syria, Ukraine, and the win that required no operation

The fourth layer is the real one, and it is not run by the INCD.

Israel’s posture toward Russia is set by a deconfliction problem. Russian forces and air defences in Syria sit between the Israeli Air Force and Iranian and Hezbollah targets, and the working relationship that permits Israeli operations there is the asset Jerusalem protects. RAND states the resulting Russian objective in the open: Moscow works to “preserve positive relations with Israel and minimize Israeli support to Ukraine“, and flags intensified Israeli activity in Syria as the friction that might strain Iranian patience with that policy (RAND 2023, pp. 15, 19).

The policy has been paid for, and one payment is documented precisely. Israel has withheld from Kyiv the cyber and intelligence assistance a country of its capability could have supplied, and its Defense Export Controls Agency blocked Ukraine from buying Pegasus and barred Estonia from using it against Russian targets in March 2022 (Feldstein and Kot, Carnegie 2023, p. 23) — a case that also shows, incidentally, that Israeli export licensing follows geopolitics rather than human-rights criteria, which is Report 2’s principle 15 failing in public.

Two conclusions, and they are the hinge of this report.

First, this is the most successful Russian operation against a first-rank cyber power on record, and it involved no operation. Moscow achieved the exclusion of one of the world’s best offensive and defensive cyber states from the coalition supporting Ukraine, at the cost of maintaining a phone line over Syria. No wiper, no intrusion, no attribution problem, no sanctions exposure. On Kerr’s distinction between activity and impact, it is the highest-impact, lowest-activity Russian success in the library.

Second, there is no technical defence against it, and pretending otherwise is how states waste money. A national SOC does not defend against a strategic bargain. The only defence is political: alignment declared publicly and in advance, so the adversary cannot buy neutrality cheaply or make each individual concession look like a reasonable act of prudence. Recorded Future’s Austrian and Czech findings show Moscow already running the same operation on a smaller budget, modulating pressure according to whether a target’s domestic politics look likely to move its way. Israel is what the finished version of that looks like.

IV · The gap: the cognitive domain

This is the section that a “learn from Israel” report has to write honestly or not write at all. On the psychological limb of information confrontation, Israel is not a model. It is a warning.

17. Nobody owns it, and the strategy says so by omission

Start with the sentence, because it comes from an Israeli institution and a former Israeli intelligence officer rather than from a foreign critic: “Israel lacks a national strategy for dealing with such foreign interventions: the Israeli Security Agency has insufficient legal powers to act, and the government’s cooperation with civil society bodies on this issue is in its infancy” (Rakov, JISS 2025, p. 9). And the comparative judgement in the same paragraph: “The threat of foreign interference… has not been given sufficient attention in Israel’s re-examination of its national priorities in the wake of the ‘Swords of Iron’ war. Israel differs in this respect from Western countries, which view the threat of disinformation and subversion with extreme severity.“

The February 2025 national cyber strategy does contain the objective — a public mindspace resilient to foreign influence sits among its eight chapters, beside “zero significant damage to critical infrastructure”, the National SOC and Cyber Dome (Shabtai, BESA 2025). What it does not contain is an owner, a legal power or a budget line — and the INCD, by design, is the wrong body to hold it. Report 2’s principle 3 explains why: the civilian authority created by Resolution 2444 was deliberately denied law-enforcement powers, “to prevent any ongoing suspicion of NSA-like practices, to build trust” (Tabansky 2020, p. 8). That choice is why Israeli companies share incident data with CERT-IL. It is also why the INCD cannot investigate a foreign influence operation, compel a platform, or act against an Israeli intermediary selling sponsored placements to an SDA front.

The Spanish military analysis of the Gaza information conflict reaches the same conclusion from outside, and its phrasing is unintentionally damning. Israel’s counter-narrative work is “likely coordinated by organizations such as the Israel Security Agency (ISA), the Ministry of Foreign Affairs, the National Security Directorate (NSD), or the government itself” (De Sebastián, IEEE 2025, p. 8). That is four candidates and a shrug — a function with no institution. Israel’s showcase response, the rapid technical rebuttal of the Al-Ahli Hospital claim with impact analyses, aerial imagery, audio and explanatory video within hours, demonstrates real capability. It also demonstrates its shape: reactive, source-based, tied to military events, and dependent on the credibility of the state making the claim.

18. Why a technically superb state is structurally blind here

Four mechanisms explain the gap, and each is a direct consequence of something that makes Israel good at the other half.

The threat model requires a named adversary and a measurable effect. Report 2’s principle 13 is that Israel specifies security against a “threat of reference”, which is why Israeli defence is concrete rather than compliance theatre. It is also why a campaign whose effects are diffuse, contested, small in any given month and unattributable to a single actor falls outside the frame. There is no threat of reference for a narrative.

The civil–military boundary that produces trust also produces a hole. The INCD cannot act; the Shin Bet lacks the powers; military intelligence is pointed outward; the Foreign Ministry does public diplomacy, not counter-interference. Every agency has a reason not to own it, and Israeli institutional design deliberately created those reasons.

The intelligence culture decommissioned the collection this problem needs. Hatzav, OSINT, document exploitation, linguistic and cultural analysis — reduced or dissolved in favour of automation before 7 October (Bar, NIPP 2024). Those are precisely the sensors that detect a Hebrew-language operation run through legally resident intermediaries.

And the political economy is hostile. Counter-disinformation in a polarised democracy is indistinguishable, to the losing side, from state regulation of domestic speech. Israel entered this period mid-way through the judicial-overhaul crisis — a fracture the campaign was explicitly designed to widen, and which Iranian operations also exploited (Freilich, INSS 2024). A government cannot credibly referee an information environment it is itself a combatant in. That is not an Israeli peculiarity; it is the general problem, and Israel is the case where it has not been solved.

19. The permissive environment, and the cost of protesting without cost

The last element is the one European states have already fixed without noticing. Rakov lists what makes Israel convenient for Russian information operations: a large number of Russian speakers, heavy air links, and “the absence of published restrictions on the activity of intelligence agents”, who “enjoy diplomatic immunity in Israel, unlike in Western countries, where there have been large-scale expulsions of Russian spies disguised as diplomats over the past five years” (p. 50). Europe’s post-2018 and post-2022 expulsions were not primarily counter-disinformation measures. They turned out to be the most effective ones available, because they removed the physical layer — the officers who stage provocations, pay intermediaries and run agents — that the digital campaign is designed to amplify.

Israel protested to Moscow in July 2023 and imposed nothing. Rakov’s recommendation is the sharpest line in the document and generalises to every European foreign ministry: “issuing warnings without imposing costs may be perceived in Moscow as positive feedback that the cognitive campaign is ‘painful’ and might portray Israel as weak” (p. 9). A démarche without a consequence is not deterrence; it is a delivery receipt.

V · Israel as an actor, not only a target

Candour here is not moralising. It is load-bearing, because it determines what Israel can credibly ask of others — and because a European state buying Israeli capability is buying from this industry.

20. The influence-for-hire industry and its lineage

The documentary record is specific. OpenAI’s October 2024 threat report discloses that “an Israel-origin commercial company we dubbed ‘Zero Zeno’ briefly generated social media comments about the elections in India, which we disrupted less than 24 hours after it began”; the earlier May 2024 disclosure attributed the network to the Israeli political-campaign firm STOIC, targeting Canadian, US, Israeli and then Indian audiences (OpenAI 2024, p. 6). Citizen Lab places this in a lineage: “both independent press and social media platforms’ investigations have previously exposed several highly sophisticated Israeli covert influence operators, and their alleged connections to the country’s government. Two notable examples are Team Jorge and Archimedes Group“ — firms that advertised prior connections to the Israeli intelligence community (Citizen Lab 2025, p. 6).

This is the flywheel Report 2 describes as the engine of Israeli cyber success, running in the other direction. Carnegie records that 80% of the 2,300 founders of Israel’s roughly 700 cyber companies served in IDF intelligence units, and the Atlantic Council’s global spyware dataset puts the Israeli cluster at 43.9% of all entities (Feldstein and Kot 2023; Atlantic Council 2024). The pipeline that produced Wiz and Check Point also produced NSO, Candiru, Intellexa — and Team Jorge. A state cannot design a pipeline that produces only the outputs it likes.

21. PRISONBREAK, and what it costs

The most consequential case is recent, forensic and hard to argue with. Citizen Lab documents a network of more than 50 inauthentic X profiles, created in 2023 and dormant until January 2025, pushing regime-change narratives at Iranian audiences and synchronised with the IDF’s June 2025 campaign. After reviewing alternatives the authors assess “that an unidentified agency of the Israeli government, or a sub-contractor working under its close supervision, is directly conducting the operation” (2025, p. 4).

The timeline is the finding. Evin Prison was struck between 11:17 and 12:18 Tehran time on 23 June 2025. A PRISONBREAK account posted about an explosion at 11:52. At 12:05, while the strikes were still under way, the network published an AI-generated video of the strike on the prison’s main entrance. The first genuine footage appeared one minute later. BBC Persian eventually flagged the video as fabricated — after international outlets had already republished it as real (pp. 6–8). The toolkit also included fabricated BBC Persian articles and bulletins the BBC confirmed it never produced, and a fully synthetic on-camera persona, “Sarina”.

The strategic cost is not squeamishness; it is operational. Israel is a documented victim of exactly this technique — the Auschwitz deepfake pushed by Doppelgänger’s European arms, the false claim on 7 October that Brigadier General Nimrod Aloni had been captured, Hamas hostage video used as an instrument of indoctrination (De Sebastián 2025, pp. 6, 15). A state that runs deniable synthetic-media operations against a foreign public forfeits the standing to demand that platforms, allies and its own citizens treat synthetic media as illegitimate — and Citizen Lab notes that attribution is already harder because platforms have restricted researcher access and cut trust-and-safety teams. Every state-linked deepfake republished by a real newsroom makes the next real atrocity easier to deny. Israel’s information operations degrade the commons that Israel’s information defence depends on. That is the sharpest available argument for the principle this library keeps arriving at: do not run covert influence operations you would condemn if aimed at you.

VI · What the evidence says about effects

Everything proposed in the final section has to survive the sceptical literature, which is the least-quoted and most important material here.

22. The effects are small, the exposure is concentrated, and the denominator is everything

The strongest study in the library is Eady and colleagues in Nature Communications, which linked a longitudinal YouGov panel of 1,496 respondents to their actual Twitter timelines during the 2016 US election. It finds no meaningful relationship between exposure to Internet Research Agency content and changes in issue positions, perceived polarisation or vote choice — coefficients near zero and, where signed, pointing the wrong way for the campaign’s aims (2023, pp. 1, 7).

Three structural facts explain it. Exposure was grotesquely concentrated: 1% of users accounted for 70% of exposures and 10% for 98%, while 1% of Russian accounts produced 89% of the content that reached timelines. The denominator is crushing: in the final month, respondents saw on average 4 IRA posts a day against 106 from national news media and 35 from politicians, and median weekly exposure was zero. And the campaign reached the least persuadable people: exposure rose monotonically with strength of Republican identification, which also sinks the “demobilise the left” theory.

The converging evidence is unusually broad. EU DisinfoLab, having discovered Doppelgänger, called it “undoubtedly a sophisticated operation, but… also a failure”, with most engagement purchased and fake pages drawing mockery of the operators’ 800-rouble bonus. OpenAI reports that “the deceptive activity that achieved the greatest social media reach and media interest was a hoax about the use of AI, not the use of AI itself”, and scored every election-related operation it disrupted at Category Two on Brookings’ six-point Breakout Scale. Elmas and colleagues, analysing 4.5 million tweets around 7 October, found 11 coordinated groups of 541 accounts whose campaigns “fail to deeply penetrate organic user networks” (2026, p. 3). Rakov puts Doppelgänger’s direct political influence in Israel at “quite limited”.

23. Coordination is not harm, and threat inflation is itself a Russian objective

Two corrections follow that matter more for institutional design than for rhetoric.

Coordination is a poor proxy for harm. Of 191 highly amplified coordinated posts in the Israel–Hamas dataset, only 12 were misleading, concentrated in 3 of the 11 groups; the rest were advocacy, religious solidarity and humanitarian mobilisation. “Treating coordination as a proxy for harm would misclassify the majority of components and risk suppressing legitimate collective action” (Elmas et al 2026, pp. 2, 28). Worse, no single signal works: claim-level integrity, toxicity and emotional tone are mutually uncorrelated. One group concentrated five misleading posts inside a fact-checking façade; another had the strongest toxicity and fear signals and zero misleading posts. Any automated early-warning or moderation system built on a single behavioural proxy fails in both directions.

Exaggeration is a deliverable. Rakov quotes Thomas Rid’s assessment that the Kremlin is Doppelgänger’s main target audience, that SDA’s activity is built around convincing the presidential administration that the campaign has serious impact, and that far more people read about Doppelgänger in the press than ever saw its content — with an explicit warning to researchers against exaggerating the significance of interference actors “and thus helping them to secure funding from their customers” (JISS 2025, p. 50). Goldstein and colleagues record the same pathology from the other side, noting that measurement ambiguity “has historically contributed to intelligence agencies inflating the impact of their influence operations for bureaucratic gain”. Eady’s residual claim closes the loop: Russia’s campaign “may have had its largest effects by convincing Americans that its campaign was successful” — the belief in interference, not the interference, is what corroded trust in the result.

Every over-stated national threat assessment is a line in a Russian contractor’s grant renewal, and a deposit in the account of domestic distrust.

24. So attack the enablers, not the narratives

None of this means doing nothing, and the deflationary findings sit beside genuinely large infrastructure numbers: 540 incidents, 10,500 channels, 34 million comments in four months, €1.56bn of state media funding, 77 Storm-1516 operations, and one deepfake of a US vice-presidential candidate viewed more than five million times on X in under 24 hours (VIGINUM 2025, p. 20). The resolution is the EEAS FIMI Deterrence Playbook, which targets three layers of the architecture — WHAT (content), HOW (digital infrastructure and supply chain), WHO (threat actors, intermediaries, contractors) — with four instruments: sanctions, law enforcement, digital regulation through the DSA and platform terms of service, and resilience-building. The logic is stated as a structural claim: “By striking at the key enablers of FIMI operations such as intermediaries, proxies and service providers the structure that sustains them can become progressively fragile” (EEAS 2026, p. 5).

The enablers are unusually European and unusually prosaic. Domain registrars where media brands are typosquatted. Hosting in Luxembourg, Estonia, Sweden and the Netherlands. A tracker registered in Estonia. Bulletproof ASNs behind DDoSia — the sanctioned Aeza International, and a UK front for a Russian host. Israeli media intermediaries selling sponsored placements. Recruited actors paid to read scripts. EU DisinfoLab’s four proposals are all of this type: protect media brands in the domain-name system, make EU-registered software and hosting liable when used for covert operations, actually enforce trademark law and the GDPR — the authors “cannot recall a single GDPR fine against the malicious actors depicted in our many investigations”, despite routine breaches — and give vetted researchers structured platform data including takedown archives.

Two operational notes complete the picture. Targeting beats volume: removing the top amplifiers of already-identified misleading posts suppressed 18.6% of misleading retweet actions at just 5% of coordinated accounts, while the naive heuristic of removing the most prolific amplifiers of all widely shared content suppressed none (Elmas et al 2026, p. 27). Prior identification of the false claims, by hand, is the binding constraint — not takedown capacity. And AI providers occupy a genuinely new vantage point in the middle of the kill chain, after an actor has accounts and access but before content is deployed; they should be treated as a reporting node, not only as a risk surface (OpenAI 2024, pp. 8–9).

VII · What this means for us

The Czech Republic is not Israel’s situation with a different flag. It is the inverse of it. Israel gets the psychological limb of Russian information confrontation without the technical limb, and has world-class technical defence and no cognitive institution. Czechia and its peers get both limbs, have a competent technical agency that is short of money and people, and have a cognitive function that has been created, abolished and relocated more than once. The transfer problem is therefore not “how do we become Israel”. It is “which half do we copy from Israel, and where do we copy the other half from”.

The Czech case is already documented, including the part that should worry most

The evidence base for Czechia is unusually complete, and it is worth setting out as one picture. The volume · 268 recorded incidents in 2024, the highest ever, but with severity falling for the third year — 1 very significant, 18 significant, 249 minor — and availability incidents alone at 43% (NÚKIB 2025, pp. 13–14). The actor · “the primary threat actor is the Russian-speaking NoName057(16), which was behind most of the DDoS attacks detected domestically”, peaking at 30 DDoS attacks in October alone, with all three Russian services present: APT29/Midnight Blizzard (SVR), COLDRIVER (FSB) and APT28 (GRU), the last with “a long-term focus on targets in the Czech Republic”. The position in the table · seventh globally at 4.00% of NoName057(16) attack days, and in the second tier of sustained ransomware victims with more than 50 attacks from Russian-affiliated groups (Brantly and Mason 2026, p. 19). The new vector · Russian-speaking hacktivists attacking weakly secured operational technology in water utilities, with one such attack registered in Czechia in 2024. The information side · the EEAS records that Russia targeted the Czech elections, with “hundreds of anonymous TikTok accounts spread[ing] pro-Kremlin narratives ahead of the parliamentary elections” (EEAS 2026, pp. 11, 14).

Two findings should be read together, because they define the actual strategic problem. First, Czechia has already acted offensively and collectively, and it did not deter anything: Military Intelligence conducted an active intervention in cyberspace against APT28 in January 2024 as part of the US-led operation DYING EMBER, and Czechia co-signed the May 2024 EU/NATO attribution of the APT28 Outlook zero-day campaign — after which, Tkachuk notes, Russia ran its combined satellite-and-cyber operation against Latvia six days later. The attribution had no deterrent effect because nothing followed it.

Second, and more uncomfortable: “the targeting throughout 2024 of the Czech Republic, a major supplier of arms to Ukraine, notably ended in January 2025 after public reporting indicated that the ruling government was expected to lose the October 2025 election to the ANO party” whose leaders had said they would halt ammunition transfers (Insikt Group 2025, p. 15). Austria received the same treatment in reverse, with attacks halting after the September 2024 election and resuming once the Freedom Party was excluded from coalition. Moscow is running on Czechia, at low cost, the same operation it has already completed against Israel: buy a policy change with graduated pressure, and reward the outcome by switching the pressure off. The Israeli case is what success looks like after a decade. It is far cheaper to refuse the bargain than to unwind it.

Against that, the binding constraint is capacity, and NÚKIB says so itself: only 47% of organisations consider their cyber budget sufficient, unchanged since 2023, while the share naming insufficient remuneration as the barrier to hiring jumped from 54% to 77% in a single year, and 55% cope by outsourcing (NÚKIB 2025, pp. 17–19). The gap is not threat awareness. It is money and people.

The models to copy are European, and they are specific

Sweden built psychological defence in 1954 as part of total defence and — unlike Norway and Denmark, which created limited wartime information councils — made it a government body active in peacetime. It lapsed into the Civil Contingencies Agency in 2002 and was re-established as a standalone Psychological Defence Agency in 2022 (Ördén, Lund 2025, p. 2). France has VIGINUM, under the prime minister’s national defence secretariat, publishing attribution-grade TLP:CLEAR technical dossiers that feed sanctions — the single most copyable institutional model in this library for a mid-sized state. The EU supplies the shared vocabulary (Pamment’s four terms: misinformation, disinformation, influence operations, foreign interference), the shared analytic grammar (ABCDE, ABC, DISARM, STIX), the regulatory lever (the DSA and the Code of Practice) and an operational centre in EEAS StratCom with a deterrence playbook already written.

Copy the safeguards with the institution. Ördén’s genealogy is the necessary warning: once information influence is treated as a permanent everyday condition rather than a wartime exception, “foreignness” becomes the criterion that licenses state action, establishing it requires intelligence and signals access, independent journalists shift from scrutineers to students, and the agency must be simultaneously transparent (for trust) and secret (for sources) — which “runs the risk of undermining its moral authority as defenders of the open democratic society”. Her constructive proposal is a governance device, not nostalgia: keep an explicit war/peace line, confine exceptional measures to wartime, and spend peacetime on the demand side — media pluralism, domestic vulnerabilities, minority- and diaspora-language media. That is cheaper and more democratic than a permanent counter-propaganda apparatus, and it is what a Czech-scale state can actually afford.

Eight moves

1. Name the owner, and give it publication powers rather than speech powers. A small counter-interference unit under the head of government on the VIGINUM model, with a statutory mandate to investigate and publish TLP:CLEAR technical reports on foreign information manipulation — and no power to order content removed. Use Pamment’s diagnostic as the routing rule: misinformation and disinformation go to education, media policy and the courts; influence operations and foreign interference go to the new unit and the intelligence services. Fifteen to twenty-five staff, not a directorate.

2. One threat picture, one attribution process, one public voice. NÚKIB’s technical assessment and the cognitive-defence function must share a single situational picture and a single quarterly national assessment, because the adversary treats them as one mission and publishes papers saying so. Keep NÚKIB’s probabilistic confidence language (”very likely (70–85%)”) across the merged product — it is a cheap, high-value practice most national agencies still lack.

3. Stand up an enabler task force and publish its scoreboard. Registrar-level protection for Czech media brands with CZ.NIC; hosting and ASN action against the bulletproof providers behind DDoSia; data-protection and trademark enforcement against cloned outlets — EU DisinfoLab cannot recall a single GDPR fine against these operators despite routine breaches; and the financial-intelligence unit on the payment rails to contractors and paid amplifiers. Report the number of actions annually, by layer: WHAT, HOW, WHO.

4. Pair every attribution with at least one imposed cost. Czechia has already attributed publicly and been ignored. Rakov’s warning generalises: a warning without a consequence reads in Moscow as confirmation that the campaign hurts. Decide in advance which instrument accompanies which finding — an expulsion, a listing, a DSA referral, a registrar takedown, a prosecution under Operation Eastwood-style coordination.

5. Buy resilience before more prevention. The Ukrainian evidence is unambiguous about what survives: universal national roaming and a decentralised operator base; legislated pre-authorisation to migrate critical state and private data to foreign clouds; mandatory backups for all critical systems; pre-arranged secure links to allied commands; and a national protective DNS, after which Ukrainian losses from financial phishing fell by more than 30% in the first month with 300-plus providers joined (Tkachuk 2025, p. 18). Every one of those is procurable now.

6. Rehearse the election playbook you already have the script for. Ukraine’s December 2021 strategic command-and-staff exercise, built from intelligence on Russian plans, saw roughly 80% of its scenario materialise within months. Czechia’s scenario is written: DDoS timed to political milestones, hundreds of anonymous TikTok accounts, deepfaked candidate audio on the Storm-1516 model, and a documented pattern of pressure being switched off when polling moves Moscow’s way. Exercise it with the electoral commission, the platforms, the broadcasters and the parties in the room.

7. Fix pay, because it is the actual constraint. A government resolution exempting a defined number of NÚKIB, national CERT and counter-interference expert posts from standard civil-service pay tables — the clause Israel wrote into Resolution 3611 in 2011, when the whole bureau’s first-year budget was NIS 4.5M. Everything else in this list fails without it.

8. Rent the sensor network while you build one. Israel’s best Russian threat picture is privately owned. A mid-sized state should procure equivalent coverage of the Russian-speaking criminal economy — dark-web forums, Telegram, infostealer logs — on the national-CERT model that vendors such as KELA now package for governments and police, with fixed sovereignty terms: source-code escrow, local hosting, open data export and clean exit. Treat it as a lease, not a capability, and keep a state analytic team able to read it.

What to copy from Israel, and what to refuse

Copy · the three-layer model with responsibility escalating by severity · a civilian defender deliberately denied police powers, which is what makes no-fault reporting work · regulation of named entities rather than whole sectors · sector SOCs feeding a national SOC · specifying defence against a named threat of reference · the continuous below-threshold “campaign between wars” mindset · and treating the domestic security industry as national capability, deliberately, with procurement to match.

Refuse · the orphaned cognitive file, and the assumption that a technical agency can hold it · an intelligence culture that decommissions OSINT, linguistic and cultural analysis because machine translation looks cheaper · an export-control regime that follows geopolitics rather than published criteria, and which blocked Ukraine from buying Pegasus and barred Estonia from using it against Russian targets · the influence-for-hire industry and the state-linked covert operations that come with it, because they destroy the commons your own defence depends on · and, above all, the bargain itself: the purchase of quiet from Moscow by withholding support from a country under attack. That is the one Israeli decision in this entire series that a European state must not copy, and it is the one Moscow would most like it to.

Close

The Russian threat has two limbs and one command. Israel has built, against a different adversary, one of the two best technical defences in the world, and it works against the Russian technical limb largely by accident. Against the psychological limb it has a strategy objective with no owner, a security service without the powers, an intelligence culture that dismantled the sensors the problem requires, and a political environment in which no government can credibly referee. Moscow noticed, and spent its effort accordingly: since April 2023 a Kremlin directorate has been running a Hebrew-language campaign with ten specialised teams and a costed plan to seat a party in the Knesset, while Russian state cyber operations went to NATO and to Ukraine.

The larger lesson is about what cyber power cannot do. Israel’s technical excellence did not prevent, and could not have prevented, the outcome Moscow actually wanted: a first-rank cyber state kept out of the coalition arming Ukraine, at the price of a deconfliction line over Syria. That result cost Russia nothing, produced no indicators, and would not have appeared in any national SOC.

For a mid-sized European state the instruction is therefore double, and neither half is optional. Build the technical machine on the Israeli pattern, because it is the best there is and most of it is affordable. And build the cognitive institution on the European pattern — Swedish in its peacetime persistence and its safeguards, French in its publication discipline, EU in its vocabulary and its enabler-focused deterrence — because Israel does not have one and knows it. Then do the cheapest and least technical thing in this report: say publicly, in advance, where you stand, so that no one has to spend a single packet finding out what it would cost to move you.