Forecasting Research Institute

Claymont, Delaware, USA

18 peopleFounded 2022

FRI advances the science of forecasting to improve decision-making on high-stakes issues including AI risk, nuclear risk, and biosecurity. It was co-founded by superforecasting pioneer Philip Tetlock.

People

Astrid Sorflaten

Head of Operations

Ezra Karger

Co-founder and Research Director

Josh Rosenberg

Chief Executive Officer

Philip E. Tetlock

President and Chief Scientist

Funding Details

Annual Budget: $2,293,916
Current Runway: -
Funding Goal: -
Funding Raised to Date: $8,687,140

Org Details

The Forecasting Research Institute (FRI) was founded in late 2022 by a team that includes Philip Tetlock — the Penn professor renowned for his work on superforecasting and the Good Judgment Project — alongside CEO Josh Rosenberg (formerly a Senior Advisor at GiveWell) and a team of researchers with backgrounds in economics, behavioral science, statistics, and public health. FRI is a 501(c)(3) nonprofit registered in Delaware and operates with roughly 15-20 staff plus a network of scientific advisors and part-time collaborators.

FRI's work is organized around two complementary tracks: foundational forecasting science and applied translational research. On the foundational side, the organization studies how to produce accurate predictions about complex, long-horizon topics — including low-probability catastrophes — and develops novel resolution methods for questions that cannot be resolved through direct observation alone. On the applied side, FRI works to make these tools actionable for organizations and policymakers.

FRI's flagship project, the Existential Risk Persuasion Tournament (XPT), ran from June to October 2022 and convened over 169 participants — including 80 superforecasters and 89 domain experts — to forecast potential catastrophic and existential risks from AI, climate change, nuclear war, and pandemics. The results were published as a peer-reviewed paper and have been widely cited in EA and policy circles.

In 2023, FRI conducted an AI Adversarial Collaboration, bringing together participants with divergent views on AI risk to identify crux questions and sharpen the debate. In 2024, FRI launched ForecastBench, a dynamic benchmark measuring LLM forecasting accuracy against superforecasters, which was updated significantly in October 2025. FRI also conducted expert studies on nuclear catastrophe risk (110 experts, 41 superforecasters) and LLM-enabled biosecurity risks (46 biosecurity experts, 22 superforecasters).

In June 2025, FRI launched LEAP (Longitudinal Expert AI Panel), a three-year project tracking monthly forecasts from over 300 AI experts, 60 superforecasters, and 1,400 public participants on AI progress indicators including benchmark performance, labor market impacts, and scientific discovery. LEAP is currently FRI's primary ongoing research initiative.

FRI is primarily funded by Coefficient Giving (formerly the Open Philanthropy forecasting portfolio) and has received multiple grants for general support and specific projects. The organization had net assets of approximately $5 million as of fiscal year 2024.

Theory of Change

FRI believes that better calibrated, quantitative forecasts about catastrophic and existential risks will improve decision-making by policymakers, philanthropists, and the broader EA community. By producing rigorous probability estimates on AI risk, nuclear risk, and biosecurity threats — and by developing the scientific methods to make such forecasts reliable — FRI creates a more accurate empirical foundation for prioritizing and funding interventions. If key decision-makers have better-calibrated uncertainty about tail risks, they can allocate resources more rationally, respond to early warning signals earlier, and avoid both over- and under-reaction to speculative threats. FRI also works to identify sources of disagreement among experts, which can surface crux questions that — if answered — would most change the field's collective beliefs about existential risk.

Grants Received

Tripwire Capability Evaluations

from Coefficient Givingforecastingresearch.org

$358,850

AI Progress Forecasting Panel

from Coefficient Givingforecastingresearch.org

$1,071,000

Forecasting Benchmark

from Coefficient Givingforecastingresearch.org

$100,000

Projects

Existential Risk Persuasion Tournament (XPT)

A 2022 forecasting tournament run by the Forecasting Research Institute that brought together 169 superforecasters and domain experts in a multi-stage deliberative process to generate high-quality long-run forecasts about existential and catastrophic risks from AI, pandemics, nuclear war, climate change, and other global threats.

active

ForecastBench

A dynamic, continuously updated benchmark developed and maintained by the Forecasting Research Institute to evaluate large language models’ forecasting accuracy on automatically generated time-series questions and prediction-market questions, comparing model performance to human superforecasters and the general public.

active

Forecasting Biosecurity Risks from LLMs

An expert elicitation study by the Forecasting Research Institute that surveys 46 biosecurity and biology experts and 22 superforecasters to estimate how frontier large language models could affect biosecurity risks and how safeguards like DNA screening and AI safety measures might mitigate those risks.

Discussion

No comments yet. Be the first to share your thoughts.