The Most Dangerous Misinformation Isn't on Social Media
We spend our energy debating misinformation on social media, where the posts are public, the platforms are accountable, and researchers can at least see what they're arguing about. The rumors that get people hurt are usually somewhere else.
A Voice Note from a Banana Stall
A couple of weeks ago in eastern Congo, four volunteers arrived at a roadside market to spray it with disinfectant. Before they could start, the admin of a local WhatsApp group recorded a voice note from his banana stall: foreign agents had come to spread Ebola, and hundreds more were being paid to infect homes along Lake Albert.
Within minutes a crowd had gathered. The volunteers were beaten, marched a mile to the next town, and handed over to police.
"We almost got lynched. We were just lucky."
— A response volunteer, to The Wall Street Journal
Where the Rumors Actually Live
Group chats are where most people encounter new information first, and it arrives from someone the group already trusts. In the DRC, it arrives as voice notes in local languages — a format that reaches people regardless of whether they read or write, and one that every text-based moderation system on earth is blind to.
The consequences the Journal documented (gift link) are specific. Patients with symptoms avoid treatment centers, fearing quarantine. Families flee when a relative dies, leaving infectious bodies for days.
One treatment center was burned down after a WhatsApp rumor claimed a burial team planned to harvest a dead man's organs. A police officer was killed trying to disperse the crowd.
This is the fastest-spreading Ebola outbreak on record: more than 2,000 cases and 754 deaths in two months, a threshold the 2018–2020 epidemic took ten months to reach. ActionAid's country director told the Journal what responders actually need, which is to "quickly track down the misleading rumors" before they turn into a mob at a health worker's back.
Meta told the Journal it is increasingly restricting forwarded content with labels and forwarding limits. That helps at the margin, but it does nothing to tell a response team in Bunia which rumor is moving through their district this morning.
Why We Built AIME
This is why we built AIME. It works where the conversation actually happens: field teams and community members forward what they're hearing into a WhatsApp-native system, and AIME takes it in whatever form it arrives, including voice notes and video in low-resource languages.
It transcribes, translates, and then does the thing no individual responder can do at volume — find the pattern across hundreds or thousands of submissions and show what's building before it breaks.
Nothing about this is passive collection. Someone chooses to send each piece of it, which is both an ethical requirement and, practically, the only way to see inside spaces that are private by design.
AIME isn't perfect, but we believe it's necessary, because response teams should have the same ground-level picture the community already has.