Summary
- The Israel Internet Association (ISOC-IL) is one registered non-profit that simultaneously holds the .il and .ישראל registry delegations, the Israeli Internet Society chapter since 1995, and a RIPE NCC registry presence — and its own 2025 documents report DNSSEC-signed .il names falling 5,861, or 88.1%, to 789.
- Three of the association's own documents attribute that fall three escalating ways: a registrar's policy change, a registrar's automatic-signing policy change, and — in the CEO's words before the board — "a failure of Israeli regulation". No registrar is named and no date is given in any of them.
- The parent .il zone never lost its DNSSEC delegation: IANA carries a secure delegation with DS 35088, last updated 2026-02-24. The collapse is in signing of child zones, which is what the registry is supposed to be able to see, and what it published without explaining.
- The commissioned premise that Israel's IPv6 adoption "stays low" does not survive the measurement record, and the article publishes the correction rather than the premise: APNIC-derived capability is 66.2% on a 2026-11-14 snapshot and 62.3% pooled for calendar 2025, against Cisco 6lab's 9.8% of users on the same page.
One non-profit, four names, three jobs
Start with who is doing the talking. IANA's root-zone record for .il lists the country-code top-level domain as Active in Latin script, managed by "The Israel Internet Association (RA)" at Petach Tikva, with Technical Status "Delegated" (IANA, .il). The same registry-of-record carries the association's second delegation: IANA's reports index records the 14 January 2021 delegation of the Hebrew-script דומיין for Israel to the Israel Internet Association (IANA reports). ICANN's country-code names supporting organisation listed the association in October 2023 as member "for the .il and the xn--4dbrk0ce / ישראל ccTLDs respectively" (ICANN ccNSO notice).
Behind the delegations sits an ordinary Israeli non-profit with an ordinary paper trail. A Registrar of Associations proper-management certificate covering 2026 gives the association name in Hebrew, association number 580299543, and states the association was registered with the Registrar on 24/03/1997; the certificate itself runs from 01/01/2026 to 31/12/2026 (Registrar certificate). It is a 2026 document reciting a 1997 date, not the original certificate, so the 1997 registration is second-hand even here.
Three dates circulate and they are not interchangeable. The association's own "who we are" page dates its founding to 1994 as an Internet interest group inside the Israeli Information Processing Association, and its registration as an association to March 1997 (ISOC-IL, who we are). The Internet Society's own newsroom says the organisation has been the Israel Chapter since 1995, and gave it a Gigabit Chapter Ranking for 2023 performance (Internet Society news). So: 1994 a working group, 1995 a chapter, 24 March 1997 a legal person. Readers who see "since 1994" on a slide and "580299543" on a certificate are looking at two different facts that do not contradict each other.
The page also explains why the same actor appears under four names in four publishers' documents. It calls itself "the Israeli branch of the global Internet Society", says it manages .il, .ישראל and the Israeli Internet Exchange, and says it represents the Israeli Internet community in ICANN, CENTR, RIPE, IETF and WSIS. That is self-description, not independent verification, and it matters that the registry's own rulebook uses yet another rendering: the Registration Rules identify the publisher as "The Internet Society of Israel (RA) (ISOC-IL)" (Registration Rules v1.10). Four renderings, one organisation, no single controlled name.
The exchange is the one part of the estate that is not held directly. The IL-IX about page states that Israel Internet Association Services Ltd., company number 517103354, is a wholly owned subsidiary of the association, and that the IIX and IL-IX exchanges are managed and operated by that subsidiary rather than by the association itself (IL-IX about). Anyone building a story about one non-profit holding everything should start by subtracting that.
What a registry actually does after it stops registering
The association's rulebook is the best available statement of what its authority consists of. Version 1.10 carries an original version of December 1998 and a stated last update of September 2025; it describes the publisher as a non-profit and as the IANA-authorised Domain Name Registry of the .il ccTLD, and it makes one operational fact unusually explicit. On 1 November 2016, ISOC-IL stopped registering and renewing domain names directly. Registration and management after that date run through ISOC-IL Accredited Registrars only.
The rules also reserve the association's right to return to acting as a registrar at any time and without prior notice.
That is the crucial mechanical point for everything that follows. After November 2016 the registry does not touch most individual names. It sets rules, accredits registrars, and counts. The act of turning DNSSEC signing on for a name belongs to whoever holds the registration relationship. When the registry later reports that signed names collapsed, it is reporting on a population it governs but does not operate.
One carve-out survived the 2016 change, and it is printed in the same document rather than left to inference: "The terms of this Section 3 will not apply with regard to the SLD .k12.il, which ISOC-IL will continue to manage directly." The exception is not a stray clause. A dedicated registry page sets out that .k12.il is intended for kindergartens and schools as classified by the Israeli Ministry of Education, that allocation happens at the fourth level in the form [name of institution].[town].k12.il, that each educational institution is entitled to only one .il domain name, and that registration under the suffix is possible only through the association via that page (special suffixes). A separate guidance page adds that any deviation from that structure needs prior approval from the Ministry of Education, and that all provisions applying to third-level names apply to fourth-level names under .k12.il (facts about domain names).
So the same organisation does two very different jobs with the same zone: a light-touch rule-maker for 289,551 names it will not touch, and a hands-on registrar for a school naming scheme with a Ministry gate. That distinction has teeth later, because "the registry" is only one of the roles in play when a signing policy changes.
The number, and the second number printed beside it
The 2025 registry data report is the primary instrument. For .il it prints 289,551 registered names at the end of 2025, up 4,793 or 1.68% year on year; 43,686 new registrations, up 3,404 or 8.45%; 38,872 deletions, down 4,266 or 9.89%; 149,488 renewals, up 2,164 or 1.47%; and 789 DNSSEC-signed names, down 5,861 or 88.1% (2025 report). For the Hebrew-script zone it prints 8,281 registered names, down 3,452 or 29.4%, with 39 DNSSEC-signed names, down 1,698 or 98%. The report's data file, published 2026-05-19, repeats the same figures (data file).
Two things about that table deserve a reader's attention. First, the report does not print the signed count as a share of the zone. Dividing 789 by 289,551 gives about 0.27%, but that is my arithmetic, not their figure, and the difference matters: a printed share is a claim, an unprinted one is an invitation.
Second, the same documents print a second, incompatible version of the headline. A figure caption in the report and its data file reads that .il changed "-1% (2,861-)" while the key table and prose of the very same documents read +1.68% (+4,793). I print both and pick no winner. That is not a detail to be smoothed away: a summary caption and the table it summarises disagreeing inside one instrument is itself a finding about how carefully the instrument is assembled, and it recurs in the companion PDF.
The prior year sets the baseline and repeats the pattern. The 2024 report prints 6,550 DNSSEC-signed names equal to 2.34% of the entire .il space at end-2024, while its key-figures block also prints 6,650 at -2% (2024 report). Two numbers 100 apart, both under the registry's name, neither labelled as a revision. Add them up and the sequence a careful reader can defend is: roughly 6,550 signed at end-2024, 789 at end-2025, with a stated cause that varies by document.
Three documents, three causes
The interesting part is not the size of the fall. It is that one organisation described it three times in three registers of formality, and the explanation grew each time.
The public data report, written for the domain-name market, gives the cause as "a policy change by one of the accredited registrars". No registrar named, no date, and no mention of a key rollover, an algorithm deprecation, a change in counting method or a new validation requirement. The sentence asks the reader to accept that a registrar did something unspecified and that this accounts for 5,861 names.
The 2025 annual activity report, submitted to the general meeting on 22 June 2026, localises the mechanism: the fall followed "a policy change concerning automatic signing of domain names at one of the association's registrars", and Table 1 records 789 signed names at -5,861 or -88.14% (annual activity report 2025). This is more useful. Automatic signing is a specific feature: a registrar can sign names on the customer's behalf by default, and a registrar can stop doing so. That moves the incident from mystery to mechanism. It still names no registrar and dates no change, and it does not say how many of the 5,861 names were being signed automatically rather than by their holders' own choice.
The board minutes of 16 April 2026 go furthest and are the most careful to attribute. They report a fall of about 5,800 signed domains owing to DNSSEC implementation policy changes at one of the association's registrars, and state that only 789 signed domains currently exist. Then the register changes: the minutes record CEO Yoram Cohen characterising the situation as "כשל של הרגולציה הישראלית" — a failure of Israeli regulation — and record that he was, as of April 2026, drafting a document surveying ten years of DNSSEC implementation in Israel (board protocol, 16.4.2026).
Read the three together and the escalation is visible. In the market-facing report a registrar did something. In the annual report the registrar stopped signing automatically. Before the board, the registry's chief executive called the underlying condition a regulatory failure. Each sentence is defensible on its own. Taken together they describe three different problems with three different owners: a commercial decision, an automation feature, and a missing legal requirement. A reader who wants to know who to hold to account is given three candidates and no way to choose.
Two limits on that reading, stated plainly. The minutes paraphrase an oral presentation; they are not a transcript, and the CEO's phrase is his characterisation, not a finding. And nothing in the record says the registrar's change fully accounts for the drop. A drop of 88% in one year is the shape of a policy cliff rather than a drift, which is consistent with the stated cause — but consistent is not the same as demonstrated.
The parent zone did not fall
Whatever happened to signing inside .il, the delegation above it held. IANA's record carries the secure-delegation block with DS 35088, algorithm 13, digest type 2, and a digest beginning ac84d274e6bf42857b41def52d8e6a23853b00fe76b7fb70b97484ed49fea182, last updated 2026-02-24. A third-party tracker puts the .il DS in the root zone at 20 July 2016 (DNSSEC deployment report).
That distinction is the difference between a bad year and a broken registry. The chain of trust from the root to .il is intact and has been since 2016; what collapsed is the number of names below it that bother to be signed. A registry can lose the first and be in genuine technical trouble. It can lose the second and still be entirely correct — because signing a child zone is a choice made by whoever runs that zone, not a duty the registry can perform on their behalf, especially after November 2016.
Which makes the missing pieces sharper. If signing is a matter of customer and registrar behaviour, then the instruments that would let an outsider judge this incident are exactly the ones a serious registry would publish: who changed what, when, which population it covered, and what the registry did about it. The 2025 report gives a number and a one-line cause. The annual report adds a mechanism. The minutes add an opinion and a promise of a ten-year retrospective then still in drafting. None of them names the registrar or fixes the date.
What is published about DNSSEC, and what is not
I looked for a registry-level DNSSEC commitment and found two documents, neither of which is one. The first is a general tutorial on Key And Signing Policy for zone operators, recommending a yearly KSK/CSK rollover and roughly 30-day signature validity as ordinary practice rather than as anything the registry commits to (policy definition page). The second is an operational notice for registrars about submitting DS records over EPP and validating DS updates against a domain's name servers (registrar portal, DNSSEC).
A DNSSEC Practice Statement — the document a registry publishes to state how it generates, stores, rolls and revokes keys, and what it guarantees to relying parties — was not found in the pages I reviewed. That is a bounded statement, not a claim about the world: several pages on the association's site returned HTTP 403 to my retrieval route, so coverage is partial and the honest form is "not obtained in this review".
What can be said is narrower and still pointed: the pages that are reachable describe how operators might sign and how registrars submit a DS, and do not describe what the registry undertakes about signing under .il.
Five numbers, three metrics, one noun
The strongest test of any registry statistic is whether anyone outside the registry publishes something comparable. Several do, and they do not agree — because they are not measuring the same thing, which is the lesson worth carrying away.
The third-party signed-name census puts .il at 1,131 signed names out of 301,021, or 0.38%, in a 14 November 2026 snapshot, with .ישראל at 21 of 6,794 (0.31%). Its method is mixed — zone files where available, HyperLogLog estimates otherwise — and its totals are approximate. Its 1,131 against the registry's 789 is a 342-name gap at different dates and methods. I treat it as an order-of-magnitude cross-check, not a refutation of either.
The Internet Society's own Pulse country report separates two things the registry's table conflates by omission. It prints coverage as "<1% of all domains under the ccTLD are protected with DNSSEC" — a supply-side share consistent with both 0.27% and 0.38% — and adoption as "31% of DNS queries are protected by DNSSEC, across all web traffic", against a 41% Asia average, while labelling .il's DNSSEC status active (Pulse, Israel). That single page is the cleanest available correction to a natural misreading: 789 unsigned names is not the same as 789 unprotected queries. Plenty of Israeli traffic is validated by resolvers even when the destination zone is not signed.
APNIC Labs measures the resolver side directly and its country series would settle the question — the page exists and is titled "Use of DNSSEC Validation for Israel (IL)" (APNIC DNSSEC, Israel). The chart is drawn in the browser and exposed no readable percentage to me; the only window I could read on the surrounding page was the global map's 30-day average for 21/10/2026 to 19/11/2026. APNIC's per-country end-client file does print a line for Israel — 2,901 unique end users, 176 behind DNSSEC-validating resolvers, 6.07% weighted, under column definitions stating the metric is the weighted percent of users whose resolvers validate (APNIC hosts file). The file carries no visible timestamp, so I do not date-anchor it, and 6.07% and Pulse's 31% are different vintages of different tabulations, not a contradiction to resolve.
Five numbers, then, for one noun: 789, 1,131, under-1%, 31% and 6.07%. Three metrics: signed names, share of domains, share of validating traffic. Only the first is the registry's own, and the registry publishes it without a denominator.
The premise that did not survive contact
The commission for this piece asked whether one non-profit holding registry authority, chapter convening and number-resource participation actually drives IPv6 deployment, "or whether the record shows the position is largely administrative while Israel's IPv6 adoption stays low". The second half of that sentence is false on the record, and it is worth saying so with the numbers rather than around them.
APNIC-derived IPv6 capability for Israel is 66.2%, with preference at 59.9%, on a 14 November 2026 snapshot built on 2,843,624 APNIC measurement samples; the same aggregator page prints Cisco 6lab's IPv6 users at 9.8%, and top-website reachability at 40.0% for web, 44.0% for email and 80.0% for DNS (stats.ipv6.army, Israel). A second aggregator draws on APNIC's daily per-economy series and prints a pooled IPv6-capable share of 62.3% for calendar 2025 against 57.5% for 2024, with a method note stating that the yearly value is IPv6-capable tests divided by all tests seen, pooled over the calendar year and unweighted within the country, unlike APNIC's user-weighted country page; it also warns that ad-delivered tests are not a random sample and that a single mobile operator turning IPv6 on can move a country by tens of points in a year (BullDozer, Israel).
66.2% and 9.8% are the same country in the same snapshot, and they are not in conflict. One is a capability share from a measurement network; the other is labelled users. Anyone quoting a single "IPv6 adoption" figure for Israel without naming the metric is not reporting, they are choosing. The honest version is that Israel's IPv6 numbers are high on APNIC's capability measure and low on Cisco's user measure, and that the gap between the two is the actual finding.
I tried to close that gap with a third independent publisher and failed in a way worth recording. The APNIC primary IPv6 country page returned only a header to me, so I attribute no primary APNIC value. Cloudflare's Radar page for Israel returned HTTP 403. Cloudflare's documentation does state that its Radar IP-version charts are percentages of HTTP requests reaching its CDN (Cloudflare Radar docs) — a different population again, not a tiebreaker.
So the commissioned question gets a corrected answer: this registry's position is not administrative, and Israel's IPv6 adoption is not low. What the record shows is narrower and more interesting. The same non-profit manages two ccTLDs, runs a school naming scheme by hand, sits in the RIPE NCC index as an LIR, and watches the signing behaviour of its zone move 88% in a year without publishing the registrar, the date, or a practice statement. That is a legibility finding, not a deployment finding.
What the chapter and the numbers can and cannot show
The address-resource half of the original frame needs the same treatment as the IPv6 half. The RIPE NCC index of "Local Internet Registries offering services in Israel" lists "Israel Internet Association" with "Registry Based in Israel" (RIPE NCC, Israel). That establishes membership and registry presence. It does not evidence a prefix or an AS number, and the same index lists organisations based elsewhere that offer services in Israel, so presence is a service claim rather than a national allocation. No allocation claim is made here.
What remains, then, is a modest and defensible set of conclusions. One registered non-profit, number 580299543, holds both Israeli ccTLD delegations and appears as a RIPE NCC LIR. It stopped registering names directly on 1 November 2016 and now governs a market of accredited registrars, with one hand-managed carve-out for schools. In 2025 its zone's signed-name count fell 88.1% to 789, and its three documents describe that fall with three different causations, none of which names the registrar or the date. Its parent delegation stayed signed throughout.
Its reachable pages describe how operators might sign and how registrars submit DS records, and do not state what the registry itself undertakes.
A final divergence to keep visible
The Hebrew-script zone closes with one more printed disagreement. The 2025 report gives 8,281 names; the 16 April 2026 board protocol gives 8,261 — a 20-name divergence at two dates and in two documents from the same organisation. Small enough to look like rounding, which is exactly why it should be printed rather than absorbed: the same body that asks readers to accept an 88.1% fall on its word is also the body whose two internal documents cannot agree on the smaller figure.
For a reader deciding how much weight to put on 789, that is the practical question. The number is plausible, cross-checked within an order of magnitude by an independent census, and consistent with a registrar-side automation change. What is missing is not the count. It is the identity, the date, and the undertaking — the three things that would turn a printed number into a record someone could act on.
Sources
- IANA, .il root-zone record
- IANA reports index
- ICANN ccNSO notice, 27 October 2023
- ISOC-IL, Registrar of Associations certificate 2026
- ISOC-IL, who we are
- Internet Society news, Gigabit Chapter Ranking
- ISOC-IL, Registration Rules v1.10
- ISOC-IL, registration under special suffixes
- ISOC-IL, facts about domain names
- ISOC-IL, 2025 registry data report
- ISOC-IL, 2025 report data file
- ISOC-IL, 2024 registry data report
- ISOC-IL, annual activity report 2025
- ISOC-IL, board protocol 16 April 2026
- ISOC-IL, DNSSEC policy definition
- ISOC-IL, registrar portal DNSSEC notice
- DNSSEC deployment report
- Internet Society Pulse, country report for Israel
- APNIC Labs, DNSSEC validation for Israel
- APNIC Labs, end clients by country
- RIPE NCC, local Internet registries offering services in Israel
- IL-IX, about
- stats.ipv6.army, Israel
- BullDozer, Israel IPv6 capability
- Cloudflare Radar, HTTP requests documentation
- BTW directory: Internet Society of Israel
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






