■ SAFE RISK ■ Public Service & Government
No. Machine analysis has become indispensable to the evidence pipeline, yet the investigation itself rests on trust built face to face with terrified people and on judgements about credibility that determine whether anyone is ever prosecuted.
“AI analyzes satellite imagery of mass graves. But interviewing survivors needs a human witness.”
Our AI replacement risk score — how we score jobs
The work splits into fieldwork and case-building. Fieldwork means getting into a refugee camp, a hospital ward or a border town, finding witnesses, establishing enough trust that a survivor will describe what happened to her family, taking testimony in a way that will survive cross-examination, photographing injuries, mapping a site, and doing all of it without exposing the witness to reprisal. Case-building means corroboration: matching accounts against satellite imagery, geolocating videos, tracing chains of command, verifying documents, and assembling a chronology that a tribunal, sanctions committee or newsroom can stand behind.
The corroboration half has been transformed. Open-source investigation now leans heavily on automated tools: change detection across satellite passes to spot razed villages or fresh earth, object recognition to identify munitions and vehicle types, shadow analysis for timestamping, face and voice matching, speech-to-text and translation across enormous volumes of social media, and clustering to find the ten relevant clips in a million uploads. Investigative units that once needed a room of analysts to review footage now triage it in hours. AI also helps detect manipulated media, which matters as denial operations get more sophisticated.
The resistant core is human on both ends. Survivors disclose to people, not interfaces, and disclosure is fragile, culturally specific and often traumatic; a badly conducted interview destroys both the witness and the evidence. Assessing credibility, spotting coached testimony, and deciding what to publish when publication may get someone killed are irreducibly judgement calls with lives attached. Courts also demand a chain of custody and a human affiant who can be cross-examined about method, and defence counsel will attack any algorithmic step that no witness can explain. Add negotiating access with hostile authorities and protecting sources, and it is clear why our score sits at 5: AI multiplies the investigator's reach without touching the mandate.
Automatability: our editorial assessment of current and near-term AI capability
Displacement is not the trajectory; reallocation is. By 2030 nearly all imagery triage, translation and media verification will be machine-first with human sign-off, meaning smaller teams cover more conflicts. Field investigators will spend proportionally more time on interviews, access negotiation and courtroom preparation. The countervailing pressure is synthetic media: as fabricated evidence improves, demand for humans who can authenticate and testify to provenance goes up, not down.
Yes, by our score among the safest. AI has absorbed the analytical grunt work, which changes the shape of teams: fewer pure researchers, more field investigators and legal specialists. Funding volatility across NGOs and UN bodies is a far bigger career risk than any model, and demand for investigators rises with each new conflict.
A great deal of the first pass. Satellite change detection, munition identification, video geolocation, translation, transcription and duplicate clustering are routinely automated, letting a small unit process volumes that were impossible a decade ago. Humans still verify each finding that will be used, because an unverified machine output is worthless in court and dangerous in a report.
It is the central emerging threat. Convincing synthetic footage lets perpetrators dismiss real atrocities as fabrications, and floods verification teams with plausible noise. The response is stronger provenance: cryptographic capture tools, documented chain of custody, and investigators who can explain in a courtroom exactly how a file was obtained and tested.
Combine three things: field interviewing with genuine trauma training, fluent use of open-source verification tools, and enough international criminal law to know what makes evidence admissible. Language skills and regional expertise still open doors nothing else does. The investigators being hired now sit at the intersection of the courtroom and the digital forensics lab.