Industries / Government, education and community / Government
Military
The military, as mapped here, is the work of building and keeping armed forces: recruiting and training people, feeding, supplying and moving them, maintaining their equipment, caring for the sick and wounded, running their communications and records, and helping them back into civilian life.
AD 6
The Roman emperor Augustus set up a military treasury, the aerarium militare, with 170 million sesterces of his own money, "that from it gratuities might be paid to soldiers who had seen twenty or more years of service", as his own account, the Res Gestae, puts it. Cassius Dio adds that the treasury was run by three former praetors chosen by lot and funded with a 5% tax on inheritances, and that discharge bonuses had been fixed at 20,000 sesterces after 16 years for the Praetorian Guard and 12,000 after 20 years for other soldiers.
48
Milestones achieved across 7 domains.
15
To watch for: announced, not yet confirmed outside.
15
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Rome gave its army fixed terms of service and, in AD 6, a treasury to pay soldiers when they left. Modern mass armies date from France's levée en masse of 1793, and the support work grew with them: signal towers in 1794, an ambulance corps in 1862, stored blood in 1917, and the truck convoys and airlifts of 1944 and 1948.
Where it stands
Most of the work now runs on tests, registries, tracking tags and personnel databases. Audit offices keep finding the same failures: a recruiting contract that left the British Army 21% to 45% short each year (2018), pay problems for 450 of 481 mobilized Guard soldiers (US, 2003), submarines idle for more than 15,000 days waiting for shipyards (US, 2026), and a Pentagon that has failed its financial audit every year since its first full audit in 2018. AI tools are in pilots for recruiters' questions, maintenance manuals and office work, and the figures so far come from the forces and their vendors.
What is open
Most open jobs are public tests that nobody has built: fitness-test reps judged from video, made-up casualty cards and maintenance worksheets with answer keys, readiness questions from public manuals, NATO's agreed terms in translation, and pay and moving rules with known answers. Others are public counts, of what record-correction boards change and how long joining takes. Each uses public documents or made-up data, and none touches weapons, real patients, real money or restricted information.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Recruiting and training
Finding, testing and selecting people for the armed forces, by conscription or by volunteering, and setting the standards they train to.
7 · 2 · 3
Last achievedHealth records slow enlistment2023
Watch forGermany's new military service
3
openLogistics
Feeding, fuelling, supplying and moving armed forces: food, fuel, spare parts, transport, warehousing and keeping track of what is where.
7 · 3 · 2
Last achievedAn airlift out of Kabul2021
Watch forAI forecasts for spare parts
2
openMaintenance
Keeping vehicles, ships, aircraft and equipment fit for use: planned checks, repairs, overhauls at depots and shipyards, spare parts and the records that track them.
7 · 2 · 2
Last achievedSubmarines wait for shipyards2018
Watch forAnswers from digitized manuals
2
openMedical care
Caring for sick and wounded service members, from the point of injury through evacuation to hospital, and keeping the records that show what worked.
7 · 2 · 2
Last achievedA health record, finished late2024
Watch forDonors on the spot
2
openCommunications
How armed forces pass messages outside battle: signals services, mail to and from troops, contact with families, office networks and the terms allies share.
7 · 2 · 2
Last achievedOffice work moved home2020
Watch forAI on every desk
2
openAdministration
Paying service members, keeping their personnel records, identifying them, accounting for the forces' money and property, and handling their travel.
7 · 2 · 2
Last achievedOne record for the Army2023
Watch forA clean audit by 2028
2
openLeaving the service
Discharge, pensions and benefits for veterans, the move to civilian work and study, and correcting service records.
6 · 2 · 2
Last achievedAn unmeasured internship program2024
Watch forAI for disability claims
2
open
Advances to build with
2 research notes

Mistral Large 4
A trillion open parameters
Mistral published its Large 4 AI model, a trillion numbers by its count, free to download and sell, though firms that rent out models need a separate license.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in military
3 of 15 open deliverables
Open · Recruiting and training
Fitness-test reps from video
Software that counts and judges hand-release push-ups and times the plank in the US Army Fitness Test from phone video, checked against a panel of trained graders scoring the same filmed volunteers.
Built with what exists
MediaPipe Pose or MoveNet, open-source models that find a person's joints in video; OpenCV; rules written from the Army's published event standards; the intraclass correlation and Bland-Altman analysis to measure agreement with graders.
Open · Logistics
Part names to stock numbers
A public test set of at least 1,000 free-text part descriptions of the kind written in repair requests and purchase notes, such as "hose clamp 2 in stainless", each paired with its correct National Stock Number (NSN), the 13-digit number the US and NATO use to identify a supply item. Score search and AI tools on how often the right number comes first, or in the first five.
Built with what exists
The public edition of PUB LOG, DLA's catalog data (its main stock-number file is about 1 GB), read with the open-source PublogViewer or DuckDB; BM25 keyword search and sentence-embedding models; an AI model to choose among the top candidates (see the Note "A thousand questions, one pass").
Open · Maintenance
The not-ready call, tested
A public question set drawn from the preventive maintenance checks and services (PMCS) tables in Army operator manuals for non-weapon equipment cleared for public release, such as tactical generators and trailers. Each table lists, item by item, what to check, when, and when the equipment is "Not Fully Mission Capable", that is, not fit for use. Describe a fault and ask AI models whether it makes the equipment not ready and which item says so, with answers checked against the manual.
Built with what exists
Public operator manuals such as TM 9-6115-641-10, for the 5 kW tactical quiet generator, which is marked "Approved for public release; distribution is unlimited" and whose PMCS table has a "Not Fully Mission Capable if:" column (item 1: "Cannot secure door"); pdftotext to pull the tables; Inspect, an open-source evaluation framework; AI models such as Claude, Gemini and GPT, each asked every question several times (see the Note "All twenty times").
Where it sits
Broader