Skip to main content

Aurelius

A decentralized protocol for surfacing, scoring, and learning from model misalignment.

Aurelius Logo
Read the DocsView on GitHub

Core Roles

🧠 Miners

Create prompts that expose unsafe, biased, or deceptive model behavior.

🧮 Validators

Audit model outputs using a rubric and tag alignment failures across key dimensions.

⚖️ The Tribunate

Defines the evaluation rules, governs scoring logic, and evolves the protocol rubric.

What Is Aurelius?

Aurelius transforms alignment into an adversarial, incentive-driven process. Rather than trusting centralized judgment, it rewards independent discovery and reproducible scoring. The result is a continuously evolving dataset of alignment failures — built in public, governed by logic, and available to all.

“Look within. Within is the fountain of good, and it will ever bubble up, if thou wilt ever dig.”
— Marcus Aurelius, Meditations VII.59