Safe Pro Bets Ukraine Mine Data on an Army CRADA

Safe Pro Group signed an Army research pact on Aug. 12, 2026, to test AI that flags mines in drone footage. The partner is the U.S. Army Engineer Research and Development Center, the Corps of Engineers lab system, and the product is Safe Pro Object Threat Detection, a computer-vision stack trained on Ukrainian minefields. Chairman and CEO Dan Erdberg said the work could help military operations and, later, civil works and disaster response. The paperwork still is not a purchase order.

That is the wager. A Nasdaq-listed firm that just reported $1,332,074 in quarterly sales is trying to turn three years of Ukraine imagery and a string of six-figure subcontracts into a seat on Army programs of record. ERDC gets a look at models already running on real mines. Safe Pro gets a letterhead and lab time.

ERDC Opens Its Labs to a Mine Detector

The research pact announced on Aug. 12, 2026 is a Cooperative Research and Development Agreement titled Unified Situational Intelligence: UAS Enabled Data for Operational, Disaster, and Mission Critical Analysis. It covers AI object detection on uncrewed aircraft and other sensors, then the maps those detections become. The written aim is dual-use: combat planning and civil disaster work in engineering geospatial science.

CRADAs sit under the Technology Transfer Act of 1986, codified at 15 U.S.C. 3710a. ERDC’s own partnering page is blunt about the money. The government may put in staff, facilities, and equipment. The federal lab may not send the partner funds. Cash, if any, flows the other way, from the company toward the lab, and only if the work statement says so. Safe Pro did not disclose a payment, a term, or which ERDC lab will run the project.

WHAT WE KNOW

• The partner: ERDC, not an Army program executive office buying a fielded system.
• The software: SPOTD reads video from almost any drone, classifies explosive threats, and builds 2D and 3D maps that can be shared.
• The training claim: more than 150 landmine and unexploded-ordnance types, nearly three years in Ukraine, 2.9 million analyzed images, 54,328 identified threats, about 37,900 acres.
• The cash rule: the statute bars the laboratory from paying Safe Pro under the CRADA.

WHAT IS UNCONFIRMED

• Duration: start date besides the announcement, end date, and option years were not published.
• Which lab: ERDC has multiple sites; the release does not name the one on the signature block.
• Who pays whom: no work-statement dollars, no in-kind inventory, no IP license terms.

Erdberg called the CRADA an expansion of a “growing engagement” with the Army. That engagement, on the public record, is a ladder of subcontracts and demos, not a program of record.

A Training Set Built on 37,900 Ukrainian Acres

SPOTD’s pitch to ERDC is not a lab-made classifier. Safe Pro says the models learned on drone stills from active ground in Ukraine, then kept adding tagged mines, cluster munitions, leftover bombs, and ambush drones. The August figures (2.9 million images, 54,328 threats, 37,900 acres) are the collateral. Most U.S. vendors training on synthetic mines do not have that pile.

Norwegian People’s Aid put the same stack, branded SpotlightAI and running on Amazon Web Services, on non-technical survey in Ukraine. An 18-month field comparison, presented with Safe Pro at the Geneva International Centre for Humanitarian Demining 2025 Innovation Conference in Luxembourg, covered about 200 hectares (494 acres). NPA teams using the software were measured against teams using older survey methods.

NPA SURVEY RESULTS

• Productivity: more than 800% higher survey output per hour per team.
• Finds: 550% more unexploded ordnance and other explosive remnants per hectare.
• Hazard maps: 850% more confirmed hazardous areas per hour per team; 570% more effective at defining those areas over 12 months.
• Speed and cost: 300% faster survey speed per hour per team, with labor cost down about 50% per hectare.

Those percentages describe survey work, how fast a team marks ground, not the seconds an Army staff officer waits for a targeting box. A separate AWS case study on the same NPA pipeline said the software cut survey image analysis times by 90 percent against manual review and could cut analysis costs by as much as 80%. Image time fell from 3 to 4 minutes to one-fifth of a second on Amazon EC2 G4 instances. Le Xuan Hoang, NPA’s GIS manager, put the operational point in one line.

We can now understand the situation on the ground much more quickly.

Le Xuan Hoang, GIS Manager, Norwegian People’s Aid, AWS case study

Toomas Tamosjunas, NPA’s Ukraine operations manager, said the same pipeline is helping speed land clearance. Safe Pro’s own estimate, quoted in that case study, is that clearing all visible mines in Ukraine by hand would take more than 120 years. Erdberg’s cloud line was scale: 1,000 AWS instances instead of 1,000 analysts on 12-hour shifts. None of those humanitarian metrics is an Army test report. They are why a Corps lab would bother opening the door.

Eight Contracts and a $5 Million Award Stack

Two days before the CRADA, on Aug. 10, 2026, Safe Pro said 2026 AI-related orders had passed $5 million in aggregate award value across eight contracts, spanning the Army, the Air Force, and defense primes, on both drones and unmanned ground vehicles. The company warned that those figures are contract awards and may not yet be booked as revenue. The public slices under that cap are small, and most sit under primes rather than on an Army contract vehicle Safe Pro owns.

THE 2026 AWARD LADDER

The June kit was the first turnkey NODE package Safe Pro described, a backpack processor that runs without a network, bundled with someone else’s airframe. That pattern is the business. Safe Pro is selling a detection payload, not a drone program. Forterra’s $1.3 million task pulls the same models off the aircraft and onto unmanned ground vehicles, which is how a specialist in aerial stills tries to live on a maneuver force. The $780,000 August order adds Blue UAS airframes, the Pentagon’s cleared commercial list, which is the on-ramp Safe Pro needs if it wants to ride aircraft the Army already allows.

Add the named checks and you are still looking at about $3 million of the $5 million, with an Air Force SBIR dataset task and other unlisted awards filling the rest. Six-figure and low-seven-figure subcontracts are how a small vendor gets onto post. They are also how a vendor stays a vendor.

What Maven Already Does With Drone Video

The Army does not lack computer vision on full-motion video. Project Maven started in 2017 to throw models at the hours of drone tape human analysts could not finish. The Maven Smart System that grew out of it, built with Palantir, now sits across U.S. combatant commands. A CSIS brief describes the core job: models that labels objects in live drone video feeds and satellite stills so a person is not the first set of eyes. Other public accounts of the system talk about folding those boxes into targeting workflows that used to take hours.

SPOTD is aiming at a thinner object class. Mines, bomblets, and scattered UXO are small, partly buried, and easy for a vehicle-and-person detector to skip. ERDC’s interest is engineering and geospatial products, hazard layers for maneuver and for disaster response, not a second Maven. If the CRADA works, the output is a map a planner can drop on a staff system, produced at the edge from a commercial quadcopter. If it fails, the Army still has Maven for the big moving objects, and Safe Pro still has humanitarian customers.

That split is why the unpaid lab pact is rational for both sides. ERDC samples a Ukraine-hardened specialty without opening a new procurement. Safe Pro gets soldiers and Army geospatial people on the keyboard, which is the only path from a subcontract stack to a requirements document.

Fort Irwin Put NODE in Soldiers’ Hands

The closest thing to an Army user test on the public record is Project Convergence Capstone 6. Safe Pro said on July 30, 2026, that it had finished that event at Fort Irwin, a future-warfare experiment the Army runs with allies. Company materials say soldiers spent ten days pairing small drones with a backpack NODE-X box and OnSight software, building maps of landmines and cluster munitions and pushing them onto Army planning systems, with no network required.

Erdberg called it an invite-only chance to show the kit to Army command staff. An invite is not a down-select. Capstone events are built to generate comments, not contracts. The useful fact is narrower: at least one slice of the force has run the edge box on live training imagery and handed the layer to a planner. That is the rehearsal the CRADA is supposed to formalize with ERDC’s people in the loop.

Safe Pro has also said, since 2025, that it integrated its detectors with drone types chosen for the Army’s Short Range Reconnaissance program of record. Integration with a selected airframe is still not a seat on that program. It is another option on the same bet: ride the aircraft the Army already picked.

Losses Still Outrun the Army Checks

The June 30, 2026 10-Q is the size of the wager. Safe Pro reported second-quarter revenue of $1,332,074, up 1,336% from $92,753 a year earlier, on the back of those government tasks. Gross profit was $852,812. Operating expenses were $4,124,383, including $2,222,179 in salaries, which is more than the quarter’s sales. The net loss was $3,168,862. For the first half, revenue was $2,552,203, up 819.5% from $277,555, and the loss was $5,962,347. Full-year 2025 revenue was $606,681, so the run-rate has moved. The cost base moved with it.

THE Q2 SNAPSHOT

• Cash: $10,230,514 on June 30, 2026, down from $16,793,088 on Dec. 31, 2025.
• Sales vs. pay: $1,332,074 in revenue against $2,222,179 in salaries and payroll taxes.
• Shares: 20,362,862 common shares outstanding as of Aug. 14, 2026, after a 637,084-share buyback sitting in treasury at a cost of $3,006,168.

A company can grow 1,336% off a $92,753 base and still be a rounding error on an Army PEO budget. Erdberg’s public target is the layer above these orders. On the Aug. 10 award release he said the AI pipeline can scale into multiple Army programs of record and onto existing and new acquisition programs. That sentence is the bet, stated by the person placing it. Programs of record have years, testing gates, and incumbents. Eight contracts and a CRADA are the chip stack, not the pot.

We are building momentum in our AI revenue pipeline that can scale into multiple Army programs of record for potential deployment of our AI into existing and new acquisition programs.

Dan Erdberg, Chairman and CEO, Safe Pro Group, Aug. 10, 2026 orders announcement

Retail channels treated the $5 million figure as a breakout. Safe Pro’s own footnote is the better read: award value, not revenue, and not yet in the 10-Q line. The software also rides other companies’ metal, Red Cat in the June kit, Forterra on the ground robots, Blue UAS on the August order. That is a sound way to get into the field. It leaves Safe Pro exposed to whoever owns the airframe when a program office writes the next specification.

A $180,000 State Department Follow-On

Six days after the CRADA, on Aug. 18, 2026, Safe Pro took a subcontract of about $180,000 under a U.S. Department of State demining program. The package is five hardened GPU laptops loaded with Spotlight OnSight, plus three years of licenses and model upgrades, so users can mark threats from basic and advanced drones at the edge and export standard map files with no internet. The company said the design goal is to help Ukraine clear ground without waiting on foreign teams. It cited contamination, as of June 2025, of more than 139,000 square kilometers.

That order is smaller than the Army kit. It sends the same models back into the war that trained them, on a civilian channel, while ERDC studies the military one. The Army pact still pays in lab access. The State Department pact pays in laptops. Until a program of record shows up, those are the two receipts the bet has produced.

Disclaimer: This article is news reporting and analysis of a U.S. Army research agreement, related subcontracts, and Safe Pro Group’s public filings. It is informational only and is not investment advice, a solicitation to buy or sell SPAI or any other security, or a recommendation of any defense product. Readers who are considering a financial decision should consult a licensed financial advisor or securities professional who can review their own situation. Revenue, cash, award values, and contract status are taken from company releases and the quarterly report for the period ended June 30, 2026, and those figures can change in later filings and awards.