Skip to main content
The Governance Error Register (GER) is an open taxonomy of AI platform governance failures. It exists so engineers, regulators, lawyers, journalists, and affected users can communicate about the same failures using the same words.

Find a code

Start here. Describe what happened — symptom picker, case search, or filter by sector — and we’ll narrow 110 codes to the 3–5 that match your case.

Browse all codes

Flat tier-grouped index. For when you already know which code you want.

Read the paper

The long-form research artifact. Methodology, related work, design principles, implications, citations.

Cite GER

Citation format, Zenodo DOI, license terms.

​
The five tiers

The GER inherits the HTTP status-code mental model. Each tier names a structural class of governance behavior:

​
What’s new in v0.2

110 codes total (up from 27 in v0.1): 108 classification codes across L1-L7 plus two Dimension Markers (GER-800, GER-900). New entries include the multi-agent failure cluster (GER-332 through GER-349), the clinical-AI cluster surfaced by Dr. Hellen V. (GER-432, GER-433), the consent and biometric cluster (GER-324, GER-329, GER-346), and refinements to the v0.1 codes with sub-type taxonomies. See the changelog for the full diff. See Attributions for the contributors whose framings shaped v0.2.

​
Two classes of codes

GER codes fall into two classes:
  • IANA-aligned codes match assigned HTTP semantics (GER-404 mirrors HTTP 404; GER-408 Timeout; GER-409 Policy Conflict; GER-500 Infrastructure Failure; GER-501 Escalation Not Implemented).
  • Namespace-claim codes occupy unassigned or extension-available HTTP slots where the failure mode has no direct HTTP analogue. This follows the convention established by Cloudflare (520–527), Twilio (1xxxx–9xxxx), and nginx (444, 494–499) in publishing domain-specific failure modes within the HTTP code space.
Both classes share the GER tier structure. The lookup tool surfaces both transparently.

​
License

GER content — codes, definitions, distinct-from boundaries, attributions, glossary, tags — is published under CC BY 4.0. Documented events are sourced from the AI Incident Database (CC BY-NC-SA 4.0). SVRNOS displays event titles and links out for full narratives — we do not redistribute the corpus.