HALO's AI may spot mines in Ukraine with about 70% accuracy

HALO announced a US$4m AWS package (25 June 2024) to trial AI on drone imagery of Ukrainian minefields; Reuters (30 April 2026) reports about 70% accuracy; no cleared land or found mine attributed.

ⓘ

AI system

System

HALO Trust AI and machine-learning analysis of drone imagery for explosive remnants of war (hosted on AWS; SageMaker used for satellite-imagery damage detection)

Model

Not assessed

Release

Not assessed

Application developer

The HALO Trust and development team with Amazon Web Services

Role

Not assessed

Configuration

Not assessed

Inputs and outputs

Not assessed

Autonomy

Not assessed

Workflow

High-resolution drone imagery of known minefields (542 flights, 11 terabytes by June 2024) manually labeled by HALO staff. AWS cloud credits, storage and SageMaker tooling to train and host the models.

Human oversight

Not assessed

Relation to AI

The model is being trained to detect explosive remnants in drone imagery with HALO analysts labeling and vetting outputs. The only performance figure, around 70% accuracy, is reported by Reuters without its basis. The organization's own June 2024 and February 2025 pages report plans and workflow, not results. No independent evaluation, cleared area or detected-mine count is documented.

AI contribution

The model is being trained to detect explosive remnants in drone imagery with HALO analysts labeling and vetting outputs. The only performance figure, around 70% accuracy, is reported by Reuters without its basis. The organization's own June 2024 and February 2025 pages report plans and workflow, not results. No independent evaluation, cleared area or detected-mine count is documented.

Benefits

Outcome identifiers

Not assessed

Benefit

achieving around 70% accuracy already

Category

Improved Task Accuracy

Status

Not assessed

Beneficiary groups

Organizations. Society.

Beneficiaries

Not assessed

Reach

Not assessed

Distribution and access

Not assessed

Duration

Not assessed

Expected?

Yes

Costs and harms

Not assessed

Measurement

Studies

Not assessed

Study design

Not assessed

Measurement

Not assessed

Setting and sample

Not assessed

Measures

achieving around 70% accuracy already

Comparison

Assumed alternative: manual analyst review of drone imagery of a minefield

Results

Assumed comparison. Reference: three to five days to look through the imagery for an average minefield.

Additional value from AI

Not assessed

Limitations

Not assessed

Denominators

Not assessed

Materials for replication

Not assessed

Money

Kind of value

Not assessed

Amount

Not assessed

Currency and prices

Not assessed

Period and unit

Not assessed

Realization

Not assessed

Recipients and payers

Not assessed

Gross or net

Not assessed

Included costs

Not assessed

Comparison

Not assessed

Method

Not assessed

Return and payback

Not assessed

People and places

Participants

Developer: The HALO Trust and development team with Amazon Web Services. Operator: The HALO Trust, Ukraine program.

Jurisdictions

Benefit: Ukraine. Developer: United Kingdom. Operator: United Kingdom.

Ownership and control

Not assessed

Dependencies

Not assessed

Cross-border connections

Not assessed

Geopolitical implications

Not assessed

Follow-up

Conditions for replication

Trained HALO personnel vet model outputs to remove false positives and negatives before survey decisions. Imagery resolution and ground conditions resemble the training minefields.

Replication attempts

Not assessed

Later findings

30 April 2026. Reuters reported from Ukraine in April 2026 that HALO's AI trained on drone imagery was identifying mines and explosive remnants with around 70% accuracy, still with no clearance outcome attributed to it.

Availability

Not assessed

Institutional decision

Not assessed

Grounds

Not assessed

Action and result

Not assessed

Success criterion

Not assessed

Evidence

Source numbers

AIBD-0315-source-0001–0003

Citations

Numbered references below.

Language

Not assessed

Source relationships

Developer, Independent

Original evidence

Not assessed

Access and versions

Not assessed

Supported findings

Not assessed

Evidence strength

Not assessed

Missing evidence

Not assessed

Grades

Assessed benefits

Organizations. Society.

Depth

Not assessed

Reasons

Organizations: HALO's pages state the aim (three to five analyst-days per minefield) and Reuters gives “around 70% accuracy”. No time saving, detection count or land release attributable to the model is documented. Society: No cited source links the AI pilot to any cleared hectare, found mine or avoided casualty in Ukraine.

Observed or intended

Not assessed

Rubric

Not assessed

Assessor and date

Not assessed

Alternative assessments

Not assessed

Fill

Independently reported

Entry details

Kind

Not assessed

Field

Humanitarian mine action. Survey and prioritization of contaminated land by a demining NGO.

Purpose

Built to flag landmines and other debris of war in drone imagery so that HALO analysts spend less time searching thousands of images per minefield. The pilot is that intended use.

Dates

Date basis unassessed: 2024 (precision unassessed). Updated: 22 October 2026.

Stage

Not assessed

Version

SHA-256: ea9458b9fb0c…

References

Other identifiers

Not assessed

Category mappings

Not assessed

Brief

Not assessed

Revisions

9 revisions. Latest: 22 October 2026.

Sources

  1. News report Independent author

    Karazy S (2026). Man, machine and AI combine in battle to clear Ukraine of mines. Reuters, 30 April 2026, syndicated by KELO.

  2. Press release Developer

    The HALO Trust (2024). AI to help accelerate landmine clearance. halotrust.org, 25 June 2024 (original URL now returns 404).

  3. Press release Developer

    The HALO Trust (2025). Using drones and AI to eradicate landmines. halotrust.org, 21 February 2025.

Collections: Cross-border cases, Security and threat-watching.