Every one of the 50 source links in the Israeli Defence-AI Deployment Index points to an English-language page. Not one grade rests on a Hebrew-language report. That is a methodological limitation with a predictable direction, and naming it is a condition of taking the index seriously: Israeli defence tech research bias toward English sourcing pushes grades down, never up.
The consequence is arithmetic. A company whose only public coverage is a Hebrew item in a domestic outlet cannot reach L1, L2 or L3 in this index, because the qualifying source was never read. It would either sit at L0 or be absent from the 50 entries altogether.
The index's sources are all in one language
The links behind the grades come from Globes' English edition, Calcalist's CTech, The Times of Israel, The Jerusalem Post, Ynetnews, Israel Defense's English site, NoCamels, Israel Hayom's English pages, and international trade press including Janes, Breaking Defense, Shephard Media, The Defense Post, Naval Today, SpaceNews and European Security & Defence.
That is a reasonable list and it is not a neutral one. Several of those Israeli outlets are English-language editions of Hebrew publications. Calcalist is a Hebrew daily business title; CTech is its English-language arm. Globes is a Hebrew-language financial daily with an English edition. Israel Defense is published in Hebrew and English editions. In each case the English edition carries a selection of what the Hebrew edition publishes, and the selection is made with an international readership in mind.
So the index is not reading Israeli defence coverage. It is reading the subset of Israeli defence coverage that an editor judged internationally interesting, plus whatever the foreign trade press picked up independently.
What the filter selects for
Editors choosing what to translate favour stories with an export angle, a US or European customer, a large funding round, or a recognisable name. They translate less often when the customer is domestic and unremarkable, when the company is small, and when the item is a routine procurement note.
Those are precisely the stories that would generate L2 grades. A modest domestic order reported in Hebrew and never translated is invisible here. The same applies to L1: a Hebrew write-up of a defence ministry evaluation, common in trade coverage, will not surface unless someone renders it into English.
The bias therefore concentrates in the middle and lower bands. It is least likely to distort L3, because reports of operational military use tend to be significant enough that some English outlet carries them. It is most likely to distort the boundary between L0 and L2.
Five of the nine L0 entries have no publicly stated founding year or location in the index at all: Aerodrome, AeroNous Solutions, Insight Intelligent Sensors, Skapion and Skyforce. A thin English-language footprint is exactly what that looks like from the outside. Whether their Hebrew footprint is equally thin is a question this index cannot answer.
Small categories are the most exposed
Israeli electronic warfare and SIGINT AI contains two entries: Cognyte, graded L2 on Israel Defense reporting, and R2 Wireless, graded L1 on Jerusalem Post reporting. Israeli military cyber AI contains three.
Both are domains where Israeli activity is widely described as substantial and where public reporting is sparse. A two-entry category is more vulnerable to a language filter than an eleven-entry one, because a single missed Hebrew report changes the distribution materially. Readers should treat the small categories as the least reliable counts in the index.
A second filter sits behind the first
Language is not the only constraint on what reaches print. Israel operates a military censor within the IDF's intelligence directorate, which conducts pre-publication review of material touching security matters. Published accounts describe roughly 2,240 press articles censored in a typical year, a small number entirely and the majority in part. CNN has reported on how international outlets operate under those arrangements during wartime, and +972 Magazine has published critical accounts of the same process.
Two effects follow for a research index. First, some domestic reporting on defence procurement and operational use is delayed, redacted or not published. Second, Israeli outlets have historically been able to carry material sourced to foreign publications that they could not report directly, which means a foreign-language original sometimes precedes the Hebrew account rather than following it.
The net direction is the same as the language filter. Both suppress the volume of publishable evidence about domestic procurement and operational use, and neither manufactures evidence that is not there. The index's central distribution, described in the deployment evidence gap, should be read as a floor.
Language bias is a known problem, not a novel one
Evidence synthesis in other fields treats this as a standard threat to validity. The Cochrane Handbook identifies language bias as a recognised source of distortion in systematic reviews. A study of Campbell Collaboration reviews found non-English studies openly excluded in more than a quarter of reviews, with only around 15 per cent including any. Research-assessment work published by DORA describes how non-indexed local-language publications become effectively invisible to evaluators.
Defence-technology research has no equivalent methodological literature, which is itself part of the problem. Borrowing the vocabulary from fields that have studied it is the cheapest available correction.
The practical fixes are unglamorous: run the search in Hebrew as well as English, treat an English edition as a sample of its Hebrew parent rather than a mirror, and record the language of every source so a reader can see the shape of the sample. Version 0.2 of this index does the third and not yet the first two.
Limits of this reading
This article cannot quantify the bias. It establishes that the sample is English-only and reasons about the likely direction, but it does not identify a single company that a Hebrew search would have graded differently, because that search has not been run.
Nor can it establish that Hebrew coverage of defence procurement is systematically richer than English coverage. It may be constrained by the same censorship and confidentiality pressures documented above, in which case the correction would be smaller than assumed. What can be stated is that the sample has a known gap, that the gap has a direction, and that a reader should discount low grades accordingly. What an index cannot tell you is treated at length in the limits of this method.
Frequently asked questions
Does the index use Hebrew-language sources?
No. All 50 grades in version 0.2 rest on English-language pages, including the English editions of Hebrew publications such as Globes, Calcalist's CTech and Israel Defense. Hebrew-only coverage was not searched, which is recorded as a limitation of the sample.
Which way does English-language bias push the grades?
Downward. A qualifying report published only in Hebrew cannot raise a grade it was never read. The effect concentrates in the contract and trial bands, where domestic orders and ministry evaluations are least likely to be translated for an international readership.
How does the Israeli military censor affect this research?
The IDF's military censor conducts pre-publication review of material touching security matters, and published accounts describe roughly 2,240 articles censored annually in whole or part. Reporting that would corroborate domestic procurement or operational use is therefore sometimes delayed, redacted or unpublished.
Sources
- idf.ai, The Israeli Defense-AI Deployment Index v1.0, 24 August 2026
- Wikipedia, Calcalist
- Wikipedia, Globes (newspaper))
- Wikipedia, Israeli Military Censor
- CNN Business, How international news outlets report under Israel's military censor during wartime, March 2026
- Cochrane Handbook 5.1, section 10.2.2.4, Language bias
- Systematic Reviews (PMC), The prevalence of and factors associated with inclusion of non-English language studies in Campbell systematic reviews
- DORA, Multilingualism and language bias in research assessment, January 2024
Independent publication of idf.ai. Not affiliated with, endorsed by, or connected to the Israel Defense Forces, the Israeli Ministry of Defense, or any government body. Compiled entirely from publicly published sources. No classified, restricted or non-public information. Listed companies may dispute any entry: send the published source that contradicts it and the entry will be amended or removed.