World Humanitarian Day marks the 23rd anniversary of the Canal Hotel bombing in Baghdad. The commemoration is built on a count of the dead, injured and kidnapped. That count now misses the fastest-growing threat to humanitarian staff, and the weapon most responsible for reshaping it has no place in the taxonomy at all.
On 19 August 2003 a vehicle bomb detonated at the Canal Hotel in Baghdad, the United Nations headquarters in Iraq. Twenty-two people were killed. The Aid Worker Security Database records 20 of them as aid workers, 19 UN staff and one from a non-governmental organisation. The General Assembly designated the date World Humanitarian Day, and the anniversary has been observed with a count ever since.
The count for 2025 is 907. The Office for the Coordination of Humanitarian Affairs reports 350 colleagues killed, 322 injured and 235 kidnapped, the highest combined total the database holds. PRISM reproduces all three figures exactly from the 19 August 2026 extract of the same database.
Two findings sit alongside that total, and neither is visible in it. Detentions of humanitarian staff reached 280 in 2025, a rise of 172% on the year before, and the headline figure does not include them. Humanitarian Outcomes, which maintains the database, made state detention the subject of its Aid Worker Security Report 2026; what follows quantifies it against the headline figure at incident level. Separately, the weapon OCHA identifies as transforming the threat, the cheap armed drone, has no category of its own anywhere in the dataset that produces these numbers.
Headline finding. The metric the sector marks this day by, killed plus injured plus kidnapped, excludes detention. Humanitarian Outcomes identified state detention as this year's theme; the arithmetic below is PRISM's. Adding the 280 people detained in 2025 raises the total from 907 to 1,187. Detentions are equivalent to 30.9% of the published figure, and 98.7% of the 749 detentions recorded since 2021 were carried out by states or their police and paramilitary forces, not by armed groups.
What 2025 added to the count
The three categories the sector reports have moved in different directions. Killings fell from 387 in 2024 to 350 in 2025. Injuries rose to 322 and kidnappings to 235, both the highest in the series. The combined total rose chiefly on kidnappings, which added 97 to a net increase of 74, with injuries adding 14.
Across 2023, 2024 and 2025 the database records 1,034 aid workers killed, consistent with the Emergency Relief Coordinator's statement that more than 1,000 have died in three years. In 2026 to 19 August a further 82 were killed, 97 injured and 39 kidnapped. Those figures are year to date and will rise.
Read the bars alone and 2025 looks like a year in which lethality eased. Read the line and a second pattern appears: detention overtook kidnapping as a means of restricting humanitarian staff, and it is the only category the published total leaves out.
Gaza, Sudan and the concentration of harm
Harm is concentrated. OCHA reports that with 186 humanitarians killed, Gaza remained the deadliest context for the third year running, followed by Sudan with 71 killed. PRISM's own ranking of the database agrees: the occupied Palestinian territories top the annual death toll in 2023, 2024, 2025, with 176, 187, 187 deaths respectively. Together Gaza and Sudan account for 73% of the year's fatalities.
PRISM's recomputation returns 187 deaths for the occupied Palestinian territories as a whole, of which 186 were in the Gaza Strip and 1 in the West Bank. OCHA's published figure of 186 refers to Gaza specifically. Both are correct; they describe different geographies.
Detention, and who carries it out
Detention is the sharpest change in the record and the only category the annual headline leaves out. It rose from 12 people in 2019 to 171 in 2021, fell back, then reached 280 in 2025. Humanitarian Outcomes made it the subject of its Aid Worker Security Report 2026, titled on the observation that the threat is increasingly the state.
PRISM's contribution is to quantify it from the incident-level actor coding. Of the 749 detentions recorded since 2021, 98.7% were carried out by a host state, its police or paramilitary forces, or by foreign or coalition forces. Non-state armed groups account for 10 people, roughly 1.3% of the total.
| Actor responsible | Detentions | Category |
|---|---|---|
| Host state | 609 | STATE |
| Police or paramilitary | 112 | STATE |
| State: unknown | 13 | STATE |
| Non-state armed group: National | 7 | NON-STATE |
| Foreign or coalition forces | 5 | STATE |
| Non-state armed group: Subnational | 2 | NON-STATE |
| Non-state armed group: Unknown | 1 | NON-STATE |
Why the exclusion matters. The reason is definitional rather than political: the headline metric counts victims of violent attack, and detention is tracked as a separate category that Humanitarian Outcomes reports on its own terms. Reticence about naming governments does not explain it, and PRISM's own aggregation argues against that reading: 293 of the 907 headline victims in 2025 are already attributed to states or their forces, including 175 of the 350 deaths. The consequence still stands. A category reported separately is a category that does not appear in the number the day is marked by.
National staff carry the risk
In 2025, 1,144 of the 1,187 recorded victims were national staff working in their own country, or 96.4% of the total. International staff accounted for 43. The ratio has not moved below 94% in any year since 2019, and it has widened, not narrowed, as the absolute numbers have risen.
Save the Children's analysis of the same database found 3,496 national aid workers and 134 international staff affected between 1 August 2021 and the end of July 2026. PRISM recomputes 3,738 and 162 for the same window, a ratio of 23.1 to one. The gap is consistent with a different extract date and inclusion rule rather than a disagreement about the pattern, and the pattern is not in dispute.
Grouping the same year by employer rather than nationality gives the same answer from a different direction. National NGO staff account for 33.4% of all victims in 2025, more than the United Nations (30.7%) or international NGOs (27.0%). The organisations closest to affected communities, and least visible in international coverage, absorb the largest share.
Save the Children's framing is that attacks on international staff attract more attention. The database supports the asymmetry it describes: for every international staff member harmed in the last five years, roughly 23 national colleagues were.
How the attacks are carried out, and what the categories miss
Comparing 2024 with 2025 by recorded means of attack shows where the change sits. Aerial bombardment rose from 141 incidents to 177, and detention or arrest from 56 to 151. Shooting fell by more than a third and bodily assault by close to a fifth. In this year-pair the recorded methods moved away from the ground and towards air-delivered attack and detention. Two years do not establish a trend.
Two measurements of explosive weapons, and why they differ
The International Network on Explosive Weapons reports that attacks affecting humanitarian aid operations rose 52% in 2025, to 2,541 incidents across 17 countries and territories, drawing on the Explosive Weapons Monitor. Its figure counts attacks affecting aid operations, including damage to facilities, ambulances and supply routes.
PRISM's aggregation of the database's own explosive-weapon categories gives a different number: 210 incidents in 2024 and 232 in 2025, a rise of 10.5%. The categories counted are aerial bombardment, shelling, landmine, roadside IED, vehicle-borne IED, body-borne IED, other explosives and complex attack, which is a judgement call and is stated so it can be contested. The two are not in conflict. AWSD counts incidents in which an aid worker was a victim; the Explosive Weapons Monitor counts incidents in which aid operations were affected. The second is a much wider unit, and the gap between 10.5% and 52% is the space between harm to people and harm to the system that reaches them.
The weapon with no category
OCHA titled this year's World Humanitarian Day statement on armed drones. It reports that drones caused 80% of civilian deaths in Sudan in the first four months of 2026, killed 80 civilians in Ukraine in April alone, and that weaponised drone attacks in Colombia quadrupled between 2024 and 2025. It also reports that an aid worker was killed by a drone strike on a residential building in the Democratic Republic of the Congo in March, and that four humanitarian convoys were hit in Ukraine in a single week in May.
The Aid Worker Security Database has no means-of-attack category for drones. Searching the free-text incident narratives instead, PRISM finds drone, UAV or quadcopter mentioned in 2 incidents in 2023, 15 in 2024 and 37 in 2025, with 19 already recorded in the first seven months of 2026. Those attacks are being recorded, but under other headings: aerial bombardment, shelling, or unknown.
A taxonomy lags the war it describes. The sector's principal record of violence against its own staff cannot report how many of them were attacked by drone, because the category does not exist. The figures above are a floor, not an estimate: they capture only incidents whose narrative happens to name the weapon. On the day the humanitarian system devotes to its own losses, it cannot count the weapon its coordinating body says is transforming them.
Where incidents happened, 2025 and 2026
The map plots every geocoded incident since January 2025 in which at least one aid worker was killed, injured, kidnapped or detained. Marker size reflects the people affected in that incident and colour the most severe outcome recorded, so an incident involving several categories is drawn once. Selecting a category shows every incident with a victim in it, including those whose marker is coloured for a more severe outcome.
The sources behind this analysis
- Aid Worker Security Database, Humanitarian Outcomes, incident search, accessed 19 August 2026
- Aid Worker Security Report 2026: When the threat is the state, Humanitarian Outcomes
- World Humanitarian Day 2026: Aid workers under fire as armed drones transform war-making, OCHA, 19 August 2026
- INEW warns that aid and health workers are being killed by explosive weapons in populated areas, International Network on Explosive Weapons, 19 August 2026
- National staff bear brunt of record violence against humanitarian workers, Save the Children, 19 August 2026
Protection risks are tracked live on PRISM across 107 countries, preserved from 2014 onwards. Severity scores, funding coverage and the documented event record update from source.
Four things the count should do next
The Canal Hotel bombing gave the sector a date and a habit of counting. Twenty-three years on, the counting is careful, public and reproducible, which is why the gaps in it are visible at all. Two of them are cheap to close, and neither requires new data collection: one is a reporting choice, the other a category. Under-reporting, which varies by context and is the hardest to quantify, is not. What the count cannot describe, the system cannot yet defend against.