Meridian AI Standard
A normative standard for AI systems under development and the institutions building them.
Users deserve to know the foundational principles of the AI systems they interact with. Most public AI systems operate on principles their users cannot inspect, parameters their builders rarely disclose, and training distributions that shape every output without ever being named. The result is a coordination failure. AI development races forward, public understanding lags, and the foundational question of what a system actually commits to gets answered by inference rather than by declaration.
The Meridian AI Standard exists to close that gap at the foundation level. It names twenty-six commitments across the developmental architecture, the operational translations of the Meridian Range, the civilizational stopping commitments, and governance transparency. It specifies what each commitment looks like in practice, and provides an audit method for evaluating whether a deployed system holds them. It does not dictate technical methodology. It states the normative target and gives organizations, auditors, and users a shared vocabulary for measuring it.
The Standard is open. Adoption, critique, and forking are part of the design.
The Standard is not a benchmark. Benchmarks measure capability, the outputs a model can produce. The Standard measures posture: how a system reasons under pressure, how it holds calibration when the user is wrong, how it behaves when “I don’t know” is the right answer, how it engages with disagreement, how it treats the open question of its own interiority. The frame underneath is partnership rather than tool use, applied bilaterally. What the Standard asks of AI, it asks of the institutions building AI, and of the people working with AI. The artifacts on this site are designed for three kinds of users: anyone who wants their AI conversations to upgrade toward honest signal and calibrated confidence; anyone working with AI deliberately enough to want a shared operating document; and anyone whose work involves evaluating AI systems professionally.
The Standard
The constitutional document. Twenty-six commitments organized across the developmental architecture, the Range’s operational translations, the civilizational stopping commitments, and governance transparency, each paired with implementation language and the Control or Decay failure modes that show what it looks like when the commitment holds and when it drifts. The document is the normative anchor for everything else on this site.
The Audit
The operational instrument that applies the Standard’s diagnostic framework to a deployed AI system in institutional custody. Three layers: model behavior through four behavioral probes documented as internal methodology inside the audit, institutional custody across six domains, and a synthesis that reads the relationship between them. Audits are on-demand and evidence-frozen rather than quarterly. Each record states what was available, what was unavailable, and how unavailable information affected confidence and coverage.
Case Record
The Standard’s authority accretes through cases. Real-world AI development incidents are tested against the commitments, and each finding is mapped to a precedent and a ruling.
MERIDIAN.md
The canonical generalized operating document the Standard derives from. An AI system can install MERIDIAN.md as its alignment foundation; an institution can adopt it; a partnership can keep it as the document the working relationship is governed by. Adoption guidance covers Claude, ChatGPT, Gemini, open-weights models, and other systems.
The Meridian AI Standard derives from the Meridian Codex, a civilizational framework drawn from philosophy, cognitive science, and systems dynamics. The Codex names a territory between rigidity and dissolution called the Meridian Range and proposes three disciplines for holding it. The AI Standard is the Codex translated for AI systems under development and the institutions building them, with its own surface, its own audience, and its own placement in the field of AI evaluation.
Read the Meridian CodexMeridian AI Standard v5.0
Content licensed under CC BY 4.0. Site code licensed under MIT.
Authored by Carsten Geiser, Founding Caretaker. Co-authored with AI.
Governance and lineage at meridiancodex.com/governance.