Skip to content
Independent research Not affiliated with, endorsed by, or connected to the Israel Defense Forces or the Israeli Ministry of Defense
idf.aiDEPLOYMENT EVIDENCE INDEX Open the index
Independent evidence indexV0.3

idf.ai — the evidence base for military AI.

idf.ai documents what defense AI has actually been fielded, and what the public record can prove. The first publication grades fifty Israeli defense-AI companies against named sources — and finds that thirty-seven of them have no published evidence of operational use.

Open the index How grades are assigned
Core principle Evidence over announcement.
EVIDENCE MAP / 2026 Israeli defense AI
Index live
The six domains the index covers and how the fifty companies fall across them ISR & SENSORS9 OF 50 AUTONOMOUS AIR7 OF 50 GROUND & SEA6 OF 50 SIMULATION5 OF 50 CYBER, EW & C212 OF 50 COUNTER-UAS11 OF 50 50 COMPANIESGRADED BY EVIDENCE
GradesL3 operational / L2 contract / L1 trial / L0 claimed
Coverage50 companies
0150 companiesIndexed and graded
0217 operationalPublic evidence of use in operations
0333 unprovenNo published operational evidence
04Every grade sourcedA working link a sceptic can open
A soldier monitors feeds on a bank of screens inside a tactical operations centre.
Photo: The U.S. Army / Public domain via Wikimedia Commons. Illustrative. Not evidence of any deployment graded on this site.

THE POSITION

Every directory of defense AI tells you who raised money. None of them tell you what has actually been fielded.

idf.ai applies one filter instead. For each company: what is the strongest piece of publicly published evidence that its product has been procured or used, and who published it. Thirty-three of the fifty companies indexed cannot clear the operational bar.

WHAT THE INDEX COVERS

Fifty companies. Six domains.

Every company in the index sits in one domain and carries one evidence grade. Each card opens the full index. The domain pages go deeper on each.

01

ISR and sensors

Imagery, signals and multi-sensor fusion, including satellite ISR. Nine companies, of which three are graded operational.

See in the index + 02

Autonomous air systems

Loitering munitions, autonomous flight and drone teaming. Seven companies, heavy on contracts and lighter on documented use.

See in the index + 03

Autonomous ground and sea

Unmanned ground vehicles, surface craft, teleoperation and off-road autonomy. Six companies.

See in the index + 04

Cyber, EW and C2

Military cyber, electronic warfare, command and control, targeting, logistics and active protection. Twelve companies.

See in the index + 05

Simulation and training

Synthetic environments and live-fire instrumentation. Five companies, graded on whether serving units actually train on them.

See in the index + 06

Counter-UAS

Detecting, tracking and defeating small drones. Eleven companies — the largest cluster in the index and the best evidenced.

See in the index +
INTERACTIVE RISK ATLAS

Risk is not a disclaimer. It is part of the system.

Select a risk to see how it appears, what to monitor and which controls belong in the lifecycle.

HUMAN FACTORS HIGH PRIORITY

Automation bias

People may accept a machine recommendation without sufficiently testing its assumptions, uncertainty or alternatives.

Signals to monitor
  • Recommendations accepted without challenge
  • Confidence displayed without context
  • Weak escalation and override patterns
Lifecycle response
  • Role-specific training and challenge procedures
  • Alternatives, provenance and uncertainty in the interface
  • Decision logs and periodic human-factors review
Research priority
05 / 05
THE EVIDENCE STANDARD

Four rules, applied to every entry.

An entry should show where a claim came from, who published it, when, and where the record runs out. These are the rules the index is built on.

01

Sources

Every grade requires a working URL a sceptical reader can open and check. No verifying source means the entry is graded L0.

02

Claims

Claims are reported, never assessed. Entries state what was published, by whom, and when — not what the author believes.

03

Flags

Company-sourced claims are flagged inside the entry, including where they are the basis for a high grade.

04

People

Founder background appears only where a named public source published it. Unit affiliations are never inferred.

EVIDENCE GRADES

Four grades. One question each.

Every company carries the highest grade its public record can support — not the highest grade it claims. Thirty-three of the fifty sit below operational.

L3Operational — 17 of 50A named public source reports the product in use in actual operations. Training exercises do not count.
L2Contract — 19 of 50A procurement contract, order or delivery is publicly documented, but use in operations is not.
L1Trial — 6 of 50A pilot, trial or evaluation is publicly reported. Nothing has entered service.
L0Claimed — 8 of 50The only source is the company itself. No independent confirmation exists in the public record.
PUBLICATIONS

One published. Three in progress.

Only the first item is live. The rest are listed so you can see where the work is going, and hold it to that.

01
DEPLOYMENT INDEX / ISRAEL

The Israeli Defense-AI Deployment Index

50 companiesLive · v1.0
02
METHOD NOTE / GRADING

Why a training exercise is not operational use

MethodNot yet published
03
ANALYSIS / THE EVIDENCE GAP

Thirty-seven of fifty: the distance between funded and fielded

AnalysisNot yet published
04
INDEX / EXPANSION

Extending the deployment index beyond Israel

IndexNot yet published
REVISIONS

One email per revision.

The index is versioned. When entries are added, regraded or corrected, subscribers get the changelog. Nothing else is sent.

No tracking beyond aggregate page statistics. One email per revision.