Case studies
Six things that actually happened, each checked against its own sources. The first ones you can act on this week. The later ones are warnings. They are here because knowing how this goes wrong is worth as much as knowing how it goes right.
Pulitzer Center: Data and Research team, Environment Investigations UnitUnited States; authors based in Kuala Lumpur and in Argentina, investigations in Brazil and Southeast Asia2025
Journalists who cannot code had an LLM write their scrapers, and published the prompts
The Pulitzer Center's data team wrote down the exact method it uses on real investigations, verbatim prompts included, together with its own warning about where the machine still fails.
- Journalist or editor
- Researcher or analyst
Read more 6 minutesFatabyyano (فتبينوا) with MeedanJordan (Amman), serving more than 19 countries across North Africa and Western Asia2022–2025
Kinan: an Arabic tipline where the machine answers "have we seen this before?" and a person answers "is it true?"
Fatabyyano and Meedan built the first Check tipline for an Arabic-speaking audience. The design decision worth copying is exactly where they drew the line between automation and human judgement.
- Communications and campaigns
- Documentation and monitoring
Read more 4 minutesAmnesty International: images posted by its Norwegian section's accountNorway / Colombia2023
Amnesty International published AI-generated images of police violence, then deleted them
The dilemma was real: how do you show an abuse without making the people in the frame traceable? The answer Amnesty reached for cost it more than the problem did. Here is the gate to walk through instead.
- Communications and campaigns
- Documentation and monitoring
Read more 4 minutesStanford University: Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu and James Zou, published in Patterns (Cell Press)United States; the finding affects second-language English writers everywhere2023
AI-text detectors flagged more than half of a set of essays written by non-native English speakers
A peer-reviewed 2023 study you can send to a donor, an editor or a university when your own writing is accused of being machine-written. Plus the four things to do before the accusation arrives.
- Researcher or analyst
- Grants and operations
Read more 4 minutesAI Hallucination Cases Database, maintained by Damien Charlotin (HEC Paris)Global; the maintainer is based in France2023–2026
A public database of court decisions in which someone filed AI-invented case citations
1,983 decisions as of 30 August 2026. You cannot build the database, but you can adopt the one-line verification rule that every entry in it documents the absence of.
- Legal work and accountability
- Researcher or analyst
Read more 4 minutesMeta Platforms; documented by Human Rights Watch, Meta's own Oversight Board, 7amleh and Access NowGlobal, focused on Palestine and Israel and on Arabic-speaking users2021–2024
How an open call for evidence and one email address produced 1,050 documented cases of content suppression
Meta's automated moderation suppressed Palestine content at scale. The part a five-person office can copy is not the technology but exactly how Human Rights Watch turned an inbox into a finding.
- Documentation and monitoring
- Communications and campaigns
Read more 6 minutes