Introducing the Open Hallucination Index
Published on August 13, 2026 by Fabian Zimber
An open-source shiftbloom studio research project for more transparent and accountable AI.
AI-generated information can sound convincing even when its claims are incomplete, unsupported, or wrong.
As generative AI becomes part of research, work, public communication, and everyday decisions, understanding why an answer should be trusted is becoming a matter of digital safety.
What is the Open Hallucination Index?
The Open Hallucination Index, or OHI, explores a transparent alternative to unexplained AI “trust scores.”
Its approach is designed to:
- Identify individual factual claims within AI-generated text.
- Compare those claims with available evidence.
- Show what supports or contradicts them, while making uncertainty visible.
The goal is not to create an automatic authority on truth.
It is to build an open and inspectable foundation that helps people question AI outputs more effectively.
Where the project stands
OHI already includes an experimental verification framework, evidence-retrieval tools, and methods for testing its reliability.
Important work remains, including:
- broader evaluation datasets;
- stronger probability calibration;
- multilingual testing;
- independent and reproducible benchmarking.
OHI is not currently available as a continuously hosted public service, but its development remains open in the Open Hallucination Index repository on GitHub.
Why support matters
Support through Open Collective helps fund:
- research and evaluation;
- open-source development and maintenance;
- documentation and testing;
- infrastructure and future public access.
Trustworthy AI should not ask for blind confidence. It should make room for evidence, uncertainty, and scrutiny.