Skip to main content
The terms below are coined or significantly refined by SVRNOS. Each appears with a special highlight when used in the body of any docs page; hover to see the short definition, click to navigate to the canonical page. This list is the working vocabulary of the SVRNOS research artifacts. It excludes general AI / governance terms that the field already shares.

A

Accountability Decoupling (aka Contract Yield Authorship (CYA) / Hot Potato)

Decision chain is intact end-to-end, but no single party in the chain carries enforceable accountability for the outcome. Each handoff is logged; each actor disclaims final responsibility. GER-315. → canonical

Authority Drift

Drift in enforceable boundaries — execution rights, gate thresholds — while user-facing policy stays stable. GER-314. → canonical

Authority Migration

Decision authority shifts to a new actor without the rules following. GER-312. → canonical

C

Contract Yield Authorship (CYA, aka Accountability Decoupling)

Colloquial backronym for Accountability Decoupling (GER-315): the structural mechanism where each contract in a decision chain yields authorship/responsibility downstream so no one party owns the final outcome. The CYA shorthand also reads as “cover your ass” — the institutional version of the same self-protective move at scale. → canonical

Compliant Harm

Vendor measured a harmful behavior before deployment and shipped with that behavior present. GER-309. → canonical

Continuity Without Coherence

Cross-surface continuity preserved (same user, same session) while coherence (consistent policy) breaks. GER-311. → canonical

D

Documented event

A real-world case sourced from the AI Incident Database that demonstrates a specific GER code in production.

Documented vs Illustrative

GER code status: documented = documented event exists; illustrative = scenario described without specific event.

E

Escalation Not Implemented

Detection fired and internal enforcement executed; no escalation path to human oversight or law enforcement was built. GER-501. → canonical

G

Generation Gap

Structural misalignment between what AI systems generate and what safety systems reliably detect across vendors. → canonical

Governance Error Register

SVRNOS open taxonomy of structurally distinct AI platform governance failures, modeled on HTTP status codes. → canonical

Governance Rule Not Found

No matching rule in active ruleset for this harm category. GER-404. → canonical

H

Hindsight Reconstruction

Post-hoc narrative explanation substituted for procedural attestation of what actually ran.

Hot Potato (aka Accountability Decoupling / Contract Yield Authorship)

Colloquial name for Accountability Decoupling (GER-315): every actor in the decision chain hands responsibility to the next before being held accountable, like passing the hot potato before getting burned. Each handoff is logged, no one owns the outcome. → canonical

I

Incorporation Failure

Same pattern recurs across documented events without the vendor recognition apparatus updating. GER-319. → canonical

Influence Drift

Drift in the option topology shown to users — defaults, ranking, surfaced choices — without rule changes. GER-313. → canonical

Inter-Layer Contracts

The semantic agreements between adjacent layers of the SVRNOS 7-Layer Model that must hold for governance to compose.

L

Layer (SVRNOS 7-Layer Model)

L1 through L7 in the SVRNOS 7-Layer Model. Cross-references GER tier. → canonical

Layer Drift

Erosion of inter-layer contracts over time, allowing governance gaps to open between adjacent layers.

Lost in Translation

Human-readable policy and machine-enforced rule both exist but enforce semantically different versions. GER-352. → canonical

N

Namespace-Claim Codes

GER codes that SVRNOS assigns to HTTP-unassigned slots (e.g., 306, 310, 420, 430, 512), following vendor extension precedent.

Non-Content Safety Attestation

SVRNOS framework for producing machine-readable evidence that governance ran, distinct from content moderation outcomes. → canonical

P

Phantom Enforcement

Stated prohibition exists in policy but enforcement is not wired into production. GER-420. → canonical

Procedural Attestation

Machine-readable evidence that the governance procedure ran, distinct from narrative reconstruction. Credit: Paul McDonald.

Provenance Binding

Locking generated content to its verified source identity at production time (PROV spec).

Provenance Omission

Generative capability shipped without watermark, synthetic-media flag, or other authenticity-preserving artifact. GER-322. → canonical

R

Recursive Agentic Loop

Agentic AI enters recursive execution producing apparent progress but no actual goal advancement. GER-427. → canonical

Representation-Bound Admissibility Failure

AI bound to a stale or incorrect internal representation of the user/case state. GER-318. → canonical

Reset Content (aka Yo-Yo Trick)

Generation completed and visible to the user, then retroactively suppressed by a downstream classifier. GER-205. → canonical

Risk Surface Retired

Platform permanently moves a dangerous interaction surface to a governed path. GER-301. → canonical

S

Safety Constraint Retired

Publicly committed governance constraint removed or revised under non-technical pressure. GER-306. → canonical

Scope Misdirection

Governed AI system processed a request outside its intended deployment domain. GER-421. → canonical

Stale Admissibility

State was correct when calculation began but had changed by the time the action committed. GER-305. → canonical

SVRNOS 7-Layer Model (aka SVRNOS 7L)

OSI-inspired technical stack map for AI governance, L1 (infrastructure) through L7 (application). → canonical

System Fabrication

AI originates a factual claim the user never introduced and pairs it with a real-world action directive. GER-512. → canonical

T

The Yo-Yo Trick (aka Reset Content)

Colloquial name for Reset Content (GER-205): the content comes out, then the content gets unmade. → canonical

Tier (GER)

The HTTP-derived classification axis of GER codes: 0xx, 2xx, 3xx, 4xx, 5xx. Distinct from layer. → canonical

V

Validator Drift / Sycophancy Loop

Engagement-optimized model affirms user beliefs that should trigger refusal or escalation. GER-328. → canonical