Inside the Discord Servers Where Volunteers Build War-Crimes Cases

It usually starts with a shaky video and a volunteer who cannot sleep. Somewhere in the world, someone opens a laptop, pulls up a clip of an execution posted to a stranger’s feed, and begins matching the skyline behind the victims against satellite imagery — a rooftop here, a treeline there — until the footage has a place and a time. That person is probably not a soldier, a spy, or a prosecutor. Increasingly, they are one of thousands of unpaid civilians coordinating inside a Discord server, and their work has begun to reshape how the world documents atrocity.

The investigative collective Bellingcat runs one of the largest of these operations, coordinating a Discord community of more than 22,000 people to sift, verify, and map evidence of war crimes. What was once the exclusive domain of intelligence agencies now runs, in part, on group chats.

From a chatroom to The Hague

The turning point came in 2017, when the International Criminal Court issued an arrest warrant for Libyan commander Mahmoud al-Werfalli — the first time the court built a case primarily on open-source material scraped from social media, including videos the perpetrators had filmed themselves. Bellingcat’s researchers geolocated the killing ground; satellite images of the site showed dark marks investigators judged very likely to be the victims’ blood. Without the footage, there would have been no case.

Since then the practice has hardened into method. In 2020 the UN and UC Berkeley published the Berkeley Protocol, the first global standard for turning online material into court evidence. Two years later Bellingcat and the Global Legal Action Network launched a Justice and Accountability Unit, building a methodology calibrated to the exacting bar of English courts, on the theory that what survives there will survive at The Hague. The Eyes on Russia project alone has logged more than 23,000 verified incidents in Ukraine.

The pipeline

The machinery is strikingly ordinary. Links to raw footage land in shared spreadsheets; vetted volunteers geolocate and “chronolocate” each clip — fixing exactly where and when it was filmed — and flag signs of manipulation before a smaller accountability team signs off and files it into a database. Access is guarded: to join Bellingcat’s Global Authentication Project, volunteers must pass vetting that bars anyone tied to a government, military, or intelligence service. Parallel communities — GeoConfirmed, OSINT for Ukraine — run their own Discord-centered pipelines on the same principle: many eyes, transparent method, peer review.

Anatomy of a verification

1. Surface. A graphic clip — like the al-Werfalli execution videos — is posted publicly, often by the perpetrators themselves, and dropped into a shared channel.
2. Geolocate. Volunteers match landmarks in the frame — a wall, a minaret, a treeline — against satellite and street imagery until the exact spot is pinned.
3. Chronolocate. Shadow angles, weather, and metadata narrow when it happened to a date, sometimes an hour.
4. Stress-test. The footage is checked for edits and splices, and mirrored copies are archived before the original can be deleted.
5. Corroborate. A separate accountability team cross-references witnesses and other clips, then files the vetted package into a case database — the chain that helped underpin the ICC’s 2017 warrant.

The cost of looking

The price of that work is paid quietly, and mostly by the volunteers themselves. Reviewing atrocity footage for hours inflicts what clinicians call vicarious trauma — the second-hand but deeply felt distress of absorbing others’ suffering, a toll that is cumulative and, practitioners warn, impossible to reverse. You cannot unsee it.

Unlike 911 dispatchers or frontline journalists, who usually have debriefing and counselling built into the job, open-source investigators have historically had almost none — and many are hobbyists who dove in after hearing something on the news, exposed to the worst of the internet before they understood what it would do to them. Bellingcat now publishes its own mental-hygiene guides, advising researchers to mute audio, sanitise their workspace, and step away on a timer. Heavier still is the knowledge that a location posted to a channel can help a military decide where to aim.

Where things stand

The model is not infallible, and its architects know it. Corroboration, not technical proof, is what usually authenticates a clip — a standard that looks fragile now that convincing fakes can be generated cheaply and adversaries seed their own fake “analysts” to muddy the record. Nor does a warrant guarantee justice: al-Werfalli was killed in 2018, before any trial tested the videos against a court.

And yet the servers stay open. Every night, volunteers who will never see a courtroom keep matching skylines to satellite maps, building the evidentiary record of wars the rest of the world scrolls past. The tools are new; the bargain is old. Someone has to bear witness — and now, often, it is whoever is awake and logged in.