Measuring truth at machine scale.
We build open methods, benchmarks, and models for determining whether information can be trusted, whether a person or a model produced it.
Explore our researchResearch areas
Provenance
Tracing where a claim originated and how it changed as it moved across platforms.
Model Evaluation
Benchmarks that test how reliably generative systems report facts.
Public Trust
Research on how people decide what to believe, and how to give them better signals.
Latest work
TruthBench: A Benchmark for Evaluating Factual Reliability in Generative Models
Model Evaluation
Signal Decay: Tracking Claim Mutation Across 4.1 Million Posts
Provenance
Calibrated Confidence: Measuring When Audiences Should Trust a Source
Public Trust
4.1M
posts analyzed in our provenance dataset
12
frontier models evaluated on TruthBench
100%
of methods published openly