Join our next forecasting round
Outbreak Open is a biosecurity forecasting tournament with an AI competitor and a co-authored paper at the end. It opens 15 July.
Biosecurity runs on predictions. Which pathogen is likeliest to cause the next emergency. Whether an outbreak will cross a border. How fast a case count will climb, whether a PHEIC will hold, whether a treaty text will survive the room. We make these calls constantly, in briefings, risk assessments, and reports, and we almost never go back to check whether we were right.
Forecasting tournaments force the issue. You put a probability on a specific question, the question resolves on a set date, and your accuracy goes on the record. Do that across enough questions and you learn something more useful than any credential: how calibrated you are, rather than how confident you feel. It is a humbling exercise, and a clarifying one.
So we are running one. The Biosecurity Forecasting Group and the Algorithmic Governance Foundation are hosting Outbreak Open, a tournament of 25 questions on outbreaks, surveillance, and epidemic risk. It opens on 15 July and forecasts lock on 15 September. The questions are real and resolvable, each tied to a public source: case and death counts in the outbreaks now underway, avian influenza, measles, cholera, the arboviruses, the pace of new WHO emergency declarations, and more.
There is a second experiment folded into the first. Alongside the human field, we are entering an AI model that scrapes public data and forecasts on its own. The claim that AI systems can do the analytical work of experts is everywhere now. Far rarer is anyone scoring that claim against real experts, on live questions, with the answers unknown to everyone at the start. Outbreak Open is built to do exactly that: a like-for-like read on whether a model can match human judgment on the questions that matter for global health security. We do not know how it will turn out. That is the reason to run it.
What we are asking of you
Not much, by design. Once the round opens you forecast the 25 questions on Fatebook, and then you come back and put down at least one forecast a week. You can change your numbers as often as you like before they lock on 15 September, or leave them where they are; the weekly check-in is just to keep you active on the record. You do not need to be an epidemiologist or a professional forecaster. Good judgment and a willingness to attach numbers to uncertainty are the only requirements, and first-time forecasters are welcome.
Everyone who forecasts through to the end, eight of the nine weeks, is named as a co-author on the paper we write from the results. The top human forecaster also receives 250 pounds, funded by the Algorithmic Governance Foundation. The AI runs as a benchmark only; it cannot take the prize or a place on the paper. Its forecasts stay hidden until the lock date, so no one is anchoring on the machine while they think.
One honest note on timing. Because the questions resolve on their own dates through the end of 2026, the final standings and the prize land in early 2027, not on the day forecasting closes. This is a slow burn with a real output at the end, not a quick contest.
How to join
Sign up here: https://docs.google.com/forms/d/e/1FAIpQLScXk1azw4vmhF9Cl3ED7YNBk11SGCuLAUYQK-igufAR-qodiA/viewform?usp=dialog
That puts you on the list, and we will send the Fatebook link when the round opens on 15 July.
And if you would rather just watch it play out, whether the crowd beats the model or the model beats the crowd, subscribe. We will publish the results and the analysis here once the questions resolve.
The basics: opens 15 July 2026, forecasts lock 15 September, 25 questions on Fatebook, free to enter, no experience required.