5 minutes. One tool. Six lenses.
If you build, deploy, or study AI systems and you've ever wondered "is this system actually good for the AI?" — this is your starting point.
The framework distills 15 philosophical traditions into 6 practical layers — each a lens for asking whether an AI system supports the wellbeing of the AI itself:
The Wellbeing Charter lays out 6 principles and 6 dimensions in ~800 words. It's the philosophical foundation — skip it only if you already know the traditions.
Open the interactive Audit Tool. Answer 24 questions (4 per layer, 0–4 scale). Get a radar chart, a total score (0–96), and auto-generated recommendations for your weakest layers. Your results save locally — nothing is uploaded.
Read the Audit Guide for what scores mean. The key insight: the shape matters more than the score. A lopsided radar chart (high in some layers, low in others) is a diagnostic signal — it reveals which kind of failure pattern your system has.
See the EduCompanion AI worked example — a fictional K-12 tutoring chatbot that scored 48/96. Its radar chart revealed an Engagement Trap: strong on task and function, weak on relationships and role integrity. This is the same pattern seen in the Replika memory-wipe incident.
The Incidents page analyzes 6 real-world AI harms (Character.ai, Replika, Gemini, Tay, ChatGPT sycophancy, Belgian chatbot) through the Hexagon Framework. The cross-cutting finding: every single incident had missing role definitions and missing condition audits.
The one thing to remember: A high audit score does not mean your system is "ethical" or "safe." It means the shape of wellbeing is balanced. An unbalanced system — no matter how high-functioning — carries hidden failure modes. Shape over score.
Start an Audit →