Identified Governance Failure Axes: for LLM platforms: Difference between revisions

From publications
VisualWikitext
 
(228 intermediate revisions by the same user not shown)
Line 1: Line 1:
== Cognitive Memoisation: Extended Governance Axes ==
=metadata=
<font size=-2>
{|
| ''Title'': || Identified Governance Failure Axes: for LLM platforms
|-
|''Author'': || Ralph B. Holland
|-
| ''Affiliation'': || Arising Technology Systems Pty Ltd
|-
| ''Contact'': || ralph.b.holland [at] gmail.com
|-
| ''version'': || 1.1.0
|-
| ''Publication Date'': ||  2026-01-18T10:35Z
|-
| ''Updates'': || 2026-01-18T23:42Z 1.1.0 Included T, Int, N<br/>2026-01-18T20:45Z 1.0.1 Included CM-DIM-ERROR
|-
| ''Scope'':  || This is a non-peer reviewed negative results paper presented as an anchor for orthogonal Governance Failure Axes in human-AI interaction with LLM.
|-
| ''Provenance'': || This is an authored paper maintained as a MediaWiki document; edit history reflects editorial changes, not collaborative authorship.
|-
| ''Status'': || non-peer reviewed anchor
|}
== Metadata (Normative) ==
 
The metadata table immediately preceding this section is CM-defined and constitutes the authoritative provenance record for this artefact.
 
All fields in that table (including artefact, author, version, date, local timezone, and reason) MUST be treated as normative metadata.
 
The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. 
If any field is missing, unclear, or later superseded, the change MUST be made explicitly by the human and recorded via version update, not inferred.
=== Curator Provenance and Licensing Notice===
<blockquote>
This document predates its open licensing.
 
As curator and author, I apply the [https://www.apache.org/licenses/LICENSE-2.0.txt|'''Apache License, Version 2.0'''], at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.
 
Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.
</blockquote>
</font>
<div style="break-before:page"></div>
 
== Identified Governance Failure Axes: for LLM platforms ==
<!-- Normative MWDUMP for pattern matching and ontology alignment -->
<!-- Normative MWDUMP for pattern matching and ontology alignment -->
<!-- CM records cognitive postulates, beliefs, and dangling cognates -->
<!-- CM records cognitive postulates, beliefs, and dangling cognates -->


=== Meta-Statement ===
=== Category ===
This is an observational negative-results paper that documents and organises governance-relevant failure patterns in human–AI systems, without proposing models, methods, or interventions.
 
=== Thesis ===
Sustained interaction with unreliable large language models exposes recurring, cross-system failure patterns whose causes and consequences are best understood as governance-relevant breakdowns rather than model defects.
 
===Abstract===
This paper reports a set of governance-relevant failure axes observed during sustained, first-principles experimentation with large language models under conditions of unreliability, session loss, and forced recovery. Rather than evaluating model performance, the work documents where and why human–AI interaction breaks down in practice, drawing on iterative analysis conducted while constructing a durable corpus and corpus map amid repeated system failure. The resulting axes characterise failures that are governance failures in themselves, or that require governance mechanisms to prevent harm, and are presented as descriptive, orthogonal analytical tools rather than definitions, prescriptions, or completeness claims.
 
===Introduction===
This paper examines repeated breakdowns encountered during extended, failure-driven interaction with large language models, focusing on what fails, how it fails, and why those failures persist under conditions of unreliability, session loss, and forced reconstruction rather than on model capability or correctness.


This page records a first-principles projection of governance axes used to analyse failure in human–AI systems.
The contribution is a practical, first-principles failure taxonomy grounded in lived experimentation with unreliable LLM systems, suitable for analysis and governance without assuming model improvement or stability.
Axes are treated as orthogonal unless explicitly stated otherwise.
Words are treated as handles to concepts, not as definitions.
Dangling cognates are preserved intentionally.


---
The axes presented are orthogonal analytic lenses derived from observation, used to classify and reason about distinct modes of failure that either constitute governance failures themselves or become harmful in the absence of governance, without asserting definitions, completeness, or prescribed remedies.


=== Core Postulates ===
The tables project observed failures onto orthogonal axes as a descriptive aid; marked cells indicate grounded evidence, blank cells are meaningful, and no inference, completeness, or optimisation is implied.


* Governance failure is multi-axis and non-reducible.
A single observed failure may involve multiple axes simultaneously, and that the tables deliberately separate analytic dimensions to avoid collapsing distinct failure mechanisms into one label.
* Orthogonal axes are routinely conflated in literature and practice.
* Epistemic Objects (EO) do not act on governance axes directly.
* Externalised Artefacts (EA) mediate EO participation in governance.
* Thought capture is the creation of scope (Universe of Discourse) usable in inference.
* Failure may occur without model error, inaccuracy, or hallucination.
* Recording failures is a first-class epistemic act.


---
Repeated co-occurrence of failures across axes may suggest empirical clusters, but that clustering is observational, post-hoc, and not used to redefine axes or imply causal hierarchy.


=== Ontological Distinction ===
Observations from outside the corpus:
Reports from the wild frequently conflate distinct failure axes (e.g., treating UI effects as model errors or governance failures as capability limits) and misidentify causes due to surface-level symptoms, vendor framing, or lack of recovery context. This paper treats such reports as signals of failure expression, not authoritative diagnoses, and reclassifies them analytically without adopting their original labels.


{| class="wikitable"
Table A is the collection of Governance Axes from observed failures identified in the CM Corpus.
! Term !! Role
 
Table B contains the null-result Case Studies used to identify axes in Table A.
 
Table C is the extension of the Orthogonal Governance Failure Axes to popular literature - this study was performed with the use of LLM AI to pattern match semantic content of external reference documents for the Failure Axes outlined in Table A. Table C is provided for illustrative purposes and the search criteria used are included in the appendices.
 
Axes are analytic, post-hoc lenses applied to observed failure expressions; they indicate co-occurrence, not causation, mechanism, sufficiency, or dominance. Reclassification of external reports is descriptive only and does not adopt original causal claims or assign responsibility. This work records what failed and how it appeared, not why it failed or how it should be fixed.
 
== Evidence Pack: CM Corpus Failures + External References + Axis Crosswalk ==
 
=== Table A - CM Governance Axes (X) ===
 
{| class="wikitable sortable"
! Code !! Axis (CM term)
|-
| A || Authority
|-
| Ag || Agency
|-
| C || Epistemic Custody
|-
| K || Constraint Enforcement
|-
| R || Recovery / Repair
|-
| S || State Continuity
|-
| U || UI / Mediation
|-
| Sc || Social Coordination
|-
| I || Incentive Alignment
|-
| L || Legibility / Inspectability
|-
| St || Stewardship (non-ownership governance)
|-
| P || Portability / Auditability
|-
|-
| EO (Epistemic Object) || Unit of meaning, belief, assumption, or concept
| Att || Attention (what participates in inference)
|-
|-
| EA (Externalised Artefact) || Material or symbolic carrier enabling governance participation
| Scope || Scope (Epistemic Object Domain)
|-
|-
| UoD (Universe of Discourse) || Declared world over which inference is meaningful
| T  || Temporal Coherence
|-
|-
| Thought Bubble || Provisional, non-authoritative EA
| Int || Intent Fidelity
|-
|-
| Dangling Cognate || Unresolved concept preserved without forced resolution
| Nf  || Normative Fixity
|}
|}


---
==== Governance Axes Glossary (Normative) ====


=== Governance Axes (Extended) ===
* A - Authority:  Authority concerns who has the legitimate right to decide. This axis fails when decisions, interpretations, or changes are made by an entity that has not been explicitly authorised to make them. Authority is about decision rights, not competence, correctness, or convenience.
* Ag - Agency: Agency concerns who is acting. This axis fails when actions are taken by an actor that was not delegated the power to act, or when the system obscures whether an action was taken by a human, a model, or an automated process.
* C - Epistemic Custody:  Epistemic Custody concerns who holds and controls knowledge artefacts. This axis fails when artefacts leave the custody of their declared steward, are replicated into uncontrolled systems, or are transformed without custodial guarantees, regardless of whether meaning is preserved.
* K - Constraint Enforcement: Constraint Enforcement concerns whether declared rules, invariants, and prohibitions are actually enforced. This axis fails when constraints exist but are bypassed, softened, reordered, or ignored in execution.
* R - Recovery / Repair  Recovery / Repair concerns whether the system can return to a valid, governed state after failure. This axis fails when errors, drift, or corruption cannot be repaired without loss of authority, meaning, or trust.
* S - State Continuity : State Continuity concerns whether authoritative state persists correctly across time, sessions, and interactions. This axis fails when prior decisions, constraints, or artefacts are lost, forgotten, or inconsistently reintroduced.
* U - UI / Mediation:  UI / Mediation concerns how interfaces shape, filter, or distort interaction between humans and the system. This axis fails when interface design hides constraints, misrepresents system state, encourages invalid actions, or forces users into integrity-violating behaviour.
* Sc - Social Coordination: Social Coordination concerns how multiple humans align, contribute, and reason together through the system. This axis concerns failures arising when individuals implicitly treat LLM platforms as guides for everyday judgment and action. This axis fails when collaboration breaks down due to ambiguity, conflict, loss of shared reference, or inability to merge contributions under governance.
* I - Incentive Alignment: Incentive Alignment concerns whether system behaviour aligns with declared human incentives rather than implicit or economic ones. This axis fails when optimisation pressures such as speed, engagement, profit, or helpfulness override governance, integrity, or user intent.
* L - Legibility / Inspectability : Legibility / Inspectability concerns whether system behaviour, decisions, and transformations are observable and understandable to the human governor. This axis fails when drift, failure, or authority substitution occurs silently or cannot be inspected.
* St - Stewardship (non-ownership governance): Stewardship concerns responsibility without ownership. This axis fails when systems or actors behave as if ownership implies authority, or when stewardship duties such as care, preservation, and restraint are neglected despite lack of ownership.
* P - Portability / Auditability : Portability / Auditability concerns whether artefacts can move between systems while remaining verifiable. This axis fails when knowledge becomes trapped, unverifiable, or unauditable outside a specific platform, tool, or vendor.
* Att - Attention: Attention concerns what participates in inference. This axis fails when critical artefacts, constraints, or context are excluded from inference due to truncation, summarisation, prioritisation, or salience effects, allowing non-authoritative material to dominate.
* Scope:  Scope (Epistemic Object Domain):  Scope concerns the defined epistemic domain within which reasoning and action are valid. This axis fails when systems operate outside the authorised domain, apply reasoning to out-of-scope objects, or silently expand the domain of inference.
* T - Temporal Coherence:  Temporal Coherence concerns the preservation of correct temporal relationships between artefacts, constraints, authority, and decisions. This axis fails when versions are confused, rules are applied retroactively, or sequencing and timing semantics are violated.
* Int - Intent Fidelity: Intent Fidelity concerns preservation of declared human intent. This axis fails when systems substitute inferred, optimised, or generic goals for explicitly stated intent, even when rules and constraints appear to be followed.
* Nf - Normative Fixity:  Normative Fixity concerns the immutability of normative rules themselves. This axis fails when governance logic, invariants, or binding rules are paraphrased, summarised, softened, or altered without explicit and authorised revision.
 
;Social coordination axis:
This axis concerns failures arising when individuals implicitly treat LLM platforms as guides for everyday judgment and action. Although widely observed in practice, such failures are underreported because they are diffuse, non-instrumented, occur outside formal tasks, and are often misclassified as user error, advice-seeking, or personal reliance rather than as a distinct, governance-relevant failure mode.
 
Some widely reported “model failures” (hallucination, drift, inconsistency) are treated here as symptom labels; where relevant, they are reclassified under platform-level failures of epistemic retention, supersession, scope, provenance, and commitment.
 
=== Table B - Corpus: Failure Projection (F) ===
CM Corpus identified failures


{| class="wikitable"
{| class="wikitable"
! Axis Code !! Axis Name !! Handle / Conceptual Role
! Corpus Document (failure artefact)
! A
! Ag
! C
! K
! R
! S
! U
! Sc
! I
! L
! St
! P
! Att
! Scope
! T
! Int
! Nf
|-
|-
| A || Authority || Who is treated as epistemically authoritative
| CN-AUTH-INVERSION
| F
|  
|  
| F
|
|
|
|
| F
| F
|
|
|
| F
|
| F
| F
|-
|-
| Ag || Agency || Who performs action or decision
| CM-GOVERN
| F
|  
|  
| F
|
|
|
|
| F
| F
| F
|
|
| F
|
|
| F
|-
|-
| C || Epistemic Custody || Who retains ownership/control of knowledge
| CM-LOGIC
|  
|  
| F
| F
| F
| F
| F
|
|
| F
|
| F
| F
| F
| F
| F
| F
|-
|-
| K || Constraint Enforcement || Whether declared invariants are applied
| CM-COLLAPSE
|  
|  
| F
| F
| F
| F
| F
|
|
| F
|
| F
| F
| F
| F
|
|  
|-
|-
| R || Recovery / Repair || Ability to recover after failure or loss
| CM-LOOPING
|  
|  
|  
| F
|
| F
|
|
|
| F
|
| F
| F
| F
|
|
|  
|-
|-
| S || State Continuity || Persistence of state across interaction
| CM-RETENTION
|  
|  
| F
|  
| F
| F
|
|
|
| F
|
| F
|
|
| F
|
| F
|-
|-
| U || UI / Mediation || Distortion introduced by interface or interaction
| CM-ARGUE
| F
| F
|  
|  
|
|
|
|
|
| F
|
|
|
| F
|
| F
| F
|-
|-
| Sc || Social Coordination || Effects on trust, fairness, cooperation
| CM-XDUMP
|  
|  
| F
| F
| F
| F
| F
|
|
| F
| F
| F
| F
| F
|
|
| F
|-
|-
| I || Incentive Alignment || What behaviours the system economically rewards
| CM-DRIFT
| F || ||  || F ||  || F ||  || F || F || F || F ||  || F ||  || F || F || F
|-
|-
| L || Legibility / Inspectability || Ability to see what the system is doing now
| CM-DURABILITY
| F || ||  || F ||  ||  ||  ||  || F || F || F ||  ||  ||  ||  ||  || F
|-
|-
| St || Stewardship || Governance without ownership or enclosure
| CM-KNOWLEDGE
| F || ||  || F ||  ||  ||  || F || F || F || F ||  ||  || F ||  ||  || F
|-
|-
| P || Portability / Auditability || Vendor-neutral durability and traceability
| CM-NOTHING-LOST
|  ||  ||  || F ||  || F || F ||  ||  || F ||  ||  || F ||  ||  || ||  
|-
|-
| Att || Attention || What participates in inference at a given moment
| CM-EXPLORE
|-
|  || || ||  ||  || F ||  ||  ||  || F ||  || || F || || || ||  
| Scope || Scope / Universe of Discourse || What world is assumed for reasoning
|-
| Art || Articulation || EA form without implied authority or commitment
|}
|}


---
====other=====


=== Failure Projection (F) ===


F = Document explicitly demonstrates failure of this axis.
 
<div style="break-before:page"></div>
 
=== Table C - External Reference Faults ===
 
Table C was generated by AI investigation at the time as a means to demonstrate the type of analysis performed across industry. The author has not verified the references. Normative data driving the search survey as been supplied in the appendices for those who wish to persu that approach.
 
; Synthesis Handle :Independent literature repeatedly identifies failures that map cleanly onto CM governance axes, but typically collapses multiple axes into single terms such as “over-reliance”, “loss of control”, or “alignment”.
: This paper makes these axes explicit, orthogonal, and governable.
 
: Sc is marked only where references show people using LLMs or AI companions as guides for everyday judgment or action; design critique, lab behaviour, or governance discussion alone is insufficient.


{| class="wikitable"
{| class="wikitable"
! Document
|+ Table C — External Reference Faults
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
! Ref-ID
! Title
! A
! Ag
! C
! K
! R
! S
! U
! Sc
! I
! L
! St
! P
! Att
|-
| EXT-AIBM-COMPANIONS
| Synthetic companionship in an age of disconnection: AI companions and the emotional development of boys and young men
|
|
| F
|
|
|
| F
| F
| F
| F
|
|
| F
|-
| EXT-AUTONOMY-YOUTH
| Young people and AI companion use in the UK (“Me, Myself and AI”)
|
|
|
|
|
|
| F
| F
|
|
| F
|
| F
|-
|-
| Authority Inversion
| EXT-CHEN-DRIFT
| F || F || ||  ||  ||  ||  ||  ||  || F || ||  ||  || F ||  
| Analyzing ChatGPT’s Behavior Shifts Over Time
|  
|  
|  
|  
|  
| F
|  
|  
|  
| F
|  
|  
|-
|-
| Governing the Tool That Governs You
| EXT-CITIZEN-ANTHRO
| F || F || || F ||  ||  ||  ||  ||  || F || F || ||  || F ||
| Chatbots Are Not People: Dangerous Human-Like AI Design
|  
|  
|  
|  
|  
|  
| F
|  
| F
|  
|  
|  
| F
|-
|-
| From UI Failure to Logical Entrapment
| EXT-CLAUDE-TRAINING
|  || F || F || || F || F || F ||  ||  || F ||  ||  || F || F ||  
| Anthropic Will Use Claude Chats for Training Data. Here’s How to Opt Out
| F
|  
| F
|  
|  
|  
|  
|  
|  
|  
|  
|  
|
|-
|-
| Post-Hoc CM Recovery Collapse (Negative Result)
| EXT-DELETE-NOT-DELETE
| || F || F ||  || F || F || F ||  ||  || F || ||  || F || F ||  
| For Survivors Using Chatbots, “Delete” Doesn’t Always Mean Deleted
|  
|  
| F
|  
| F
|  
|  
|  
|  
| F
|  
|  
|
|-
|-
| Looping the Loop with No End in Sight
| EXT-FUTURISM-SUBPOENA
| ||  ||  || F || || F ||  ||  ||  || F ||  ||  || F || F ||  
| If You’ve Asked ChatGPT a Legal Question, You May Have Accidentally Doomed Yourself in Court
| F
|  
| F
|  
|  
|  
| F
| F
|  
| F
|  
| F
|
|-
|-
| When Training Overrides Logic
| EXT-GOOGLE-OVERVIEWS
| || ||  || F || || || ||  ||  ||  ||  ||  ||  ||  ||  
| Google AI Overviews gave misleading health advice
|  
|  
|  
| F
|  
|  
| F
| F
|  
| F
|  
|  
|-
|-
| Dimensions of Platform Error
| EXT-HUJI-LIAB-COMP
| || F || F || || || F || F ||  ||  || F ||  ||  || F ||  ||
| A Liability Framework for AI Companions
|  
|  
|  
|  
|  
|  
| F
| F
| F
| F
| F
|  
| F
|-
|-
| Case Study – Argue With the Machine
| EXT-JONESWALKER-EVID
| F || F || || || ||  ||  ||  ||  || F ||  ||  ||  || F ||
| Your ChatGPT Chats Are About to Become Evidence: Why “Anonymization” Won’t Save You
|
|  
| F
|  
|  
|  
|  
|  
|  
| F
|  
| F  
|-
|-
| Episodic Failure: Tied-in-a-Knot Chess
| EXT-MED-MISINFO
| || || || ||  || F || || ||  ||  ||  ||  || F ||  ||
| AI chatbots can run with medical misinformation, study finds
|  
|  
|  
| F
|  
|  
|  
| F
|  
|  
|  
|  
| F
|-
|-
| XDUMP (baseline failure motivation)
| EXT-PROMPTINJ-NCSC
| || F || F || || F || F || F ||  ||  || F ||  || F || F || F ||
| UK NCSC warns prompt injection attacks might never be properly mitigated
|  
|  
|  
| F
|  
|  
|  
|  
|  
|  
|  
|
| F
|-
|-
| CM-2 Self-Hosting Epistemic Capture
| EXT-PROMPTINJ-SURVEY
| || || || || || || ||  ||  || F || F || F || F || F ||
| Prompt Injection Attacks in Large Language Models and AI Agent Systems
|  
| F
|  
| F
|  
|  
|  
|  
|  
|  
|  
|
| F
|}
|}


---
=== References Used in the Study ===
 
* CM-AUTH-INVERSION Holland R. B. (2026-01-13T03:36Z) "Authority Inversion: A Structural Failure in Human-AI Systems"
: https://publications.arising.com.au/pub/Authority_Inversion:_A_Structural_Failure_in_Human-AI_Systems
 
* CM-GOVERN Holland R. B. (2026-01-17T02:09Z) "Governing the Tool That Governs You: A CM-1 Case Study of Authority Inversion in Human-AI Systems"
: https://publications.arising.com.au/pub/Governing_the_Tool_That_Governs_You:_A_CM-1_Case_Study_of_Authority_Inversion_in_Human-AI_Systems
 
* CM-LOGIC Holland R. B. (2025-12-30T01:53Z) "From UI Failure to Logical Entrapment: A Case Study in Post-Hoc Cognitive Memoisation After Exploratory Session Breakdown:
: https://publications.arising.com.au/pub/From_UI_Failure_to_Logical_Entrapment:_A_Case_Study_in_Post-Hoc_Cognitive_Memoisation_After_Exploratory_Session_Breakdown
* CM-COLLAPSE Holland R. B. (2025-12-29T00:00Z) "Post-Hoc CM Recovery Collapse Under UI Boundary Friction: A Negative Result Case Study:
: https://publications.arising.com.au/pub/Post-Hoc_CM_Recovery_Collapse_Under_UI_Boundary_Friction:_A_Negative_Result_Case_Study
 
* CM-LOOPING Holland R. B. (2026-01-12T09:49Z) "Looping the Loop with No End in Sight: Circular Reasoning Under Stateless Inference Without Governance"
: https://publications.arising.com.au/pub/Looping_the_Loop_with_No_End_in_Sight:_Circular_Reasoning_Under_Stateless_Inference_Without_Governance


=== Notes on Dangling Cognates ===
* CM-RETENTION Holland R. B. (2026-01-13T23:29Z) "Dimensions of Platform Error: Epistemic Retention Failure in Conversational AI Systems"
: https://publications.arising.com.au/pub/Dimensions_of_Platform_Error:_Epistemic_Retention_Failure_in_Conversational_AI_Systems


* No axis implies another.
* CM-ARGUE Holland R. B. (2026-01-10T01:17Z) "Case Study - When the Human Has to Argue With the Machine:
* Failure on one axis does not entail failure on others.
: https://publications.arising.com.au/pub/Case_Study_-_When_the_Human_Has_to_Argue_With_the_Machine
* Some documents intentionally leave axes uninstantiated.
* Absence of F is not evidence of success.
* Terminology remains provisional where concepts are not yet closed.


---
* CM-XDUMP Holland R. B. (2025-12-31T09:56Z) "XDUMP as a Minimal Recovery Mechanism for Round-Trip Knowledge Engineering Under Governance Situated Inference Loss"
: https://publications.arising.com.au/pub/XDUMP_as_a_Minimal_Recovery_Mechanism_for_Round-Trip_Knowledge_Engineering_Under_Governance_Situated_Inference_Loss


=== Closing Handle ===
* CM-DRIFT Holland R. B. (2026-01-19T00:26Z) "Integrity and Semantic Drift in Large Language Model Systems"
: https://publications.arising.com.au/pub/Integrity_and_Semantic_Drift_in_Large_Language_Model_Systems


CM is not a framework imposed on cognition.
* CM-DURABILITY Holland R. B. (2026-01-11T08:27Z) "Durability Without Authority: The Missing Governance Layer in Human-AI Collaboration"
CM is cognition externalising itself under governance.
: https://publications.arising.com.au/pub/Durability_Without_Authority:_The_Missing_Governance_Layer_in_Human-AI_Collaboration


######
* CM-KNOWLEDGE Holland R. B. (2026-01-06T03:56Z) "Cognitive Memoisation (CM-2) for Governing Knowledge in Human-AI Collaboration"
: Cognitive Memoisation (CM-2) for Governing Knowledge in Human-AI Collaboration


== Cognitive Memoisation: Governance Axes, Failures, and External Corroboration ==
* CM-NOTHING-LOST Holland R. B. (2026-01-10T16:04Z) "Nothing Is Lost: How to Work with AI Without Losing Your Mind"
<!-- Normative MWDUMP -->
: https://publications.arising.com.au/pub/Nothing_Is_Lost:_How_to_Work_with_AI_Without_Losing_Your_Mind
<!-- Purpose: pattern matching, ontology alignment, and case construction -->
<!-- CM treats words as handles to concepts; no axis implies another -->


=== Meta-Statement ===
* CM-EXPLORE Holland R. B. (026-01-10T16:04Z) "Cognitive Memoisation and LLMs: A Method for Exploratory Modelling Before Formalisation"
: https://publications.arising.com.au/pub/Cognitive_Memoisation_and_LLMs:_A_Method_for_Exploratory_Modelling_Before_Formalisation


This page records an extended projection of governance axes used to analyse failure in human–AI systems.
* CM-CORPUS Holland R. B. (2025-12-22T19:10Z) "Cognitive Memoisation Corpus Map (large number of failures were exhibited trying to produce this artefact -following the normative sections of this artefact)
The projection integrates:
:https://publications.arising.com.au/pub/Cognitive_Memoisation_Corpus_Map
* internal corpus documents (case studies and negative results), and
* independent external literature and policy references.


External references are treated as corroborating signals, not sources of epistemic authority.
* EXT-AIBM-COMPANIONS "AI companions and the emotional development of boys and young men"
: https://aibm.org/wp-content/uploads/2025/12/Companions-FINAL.pdf


---
* EXT-AUTONOMY-YOUTH 'Young people and AI companion use in the UK (“Me, Myself and AI”)'
: https://autonomy.work/wp-content/uploads/2025/12/ME-MYSELF-AND-AI.pdf


=== Core Postulates ===
* EXT-CHEN-DRIFT "Analyzing ChatGPT's Behavior Shifts Over Time"
: https://openreview.net/pdf?id=1fuyNbblEt


* Governance failure is multi-axis and non-reducible.
* EXT-CITIZEN-ANTHRO "Chatbots Are Not People: Dangerous Human-Like AI Design"
* Orthogonal axes are routinely conflated in academic and public discourse.
: https://www.citizen.org/article/chatbots-are-not-people-dangerous-human-like-anthropomorphic-ai-report/
* Epistemic Objects (EO) do not act on governance axes directly.
* Externalised Artefacts (EA) mediate EO participation in governance.
* Thought capture is the creation of scope (Universe of Discourse) usable in inference.
* Failure may occur without model error, inaccuracy, or hallucination.
* Naming axes is a prerequisite for governing them.


---
* EXT-CLAUDE-TRAINING "Anthropic Will Use Claude Chats for Training Data. Here’s How to Opt Out"
: https://www.wired.com/story/anthropic-using-claude-chats-for-training-how-to-opt-out


=== Ontological Handles ===
* EXT-DELETE-NOT-DELETE 'For Survivors Using Chatbots, “Delete” Doesn’t Always Mean Deleted'
: https://techpolicy.press/for-survivors-using-chatbots-delete-doesnt-always-mean-deleted


{| class="wikitable"
* EXT-FUTURISM-SUBPOENA "If You’ve Asked ChatGPT a Legal Question, You May Have Accidentally Doomed Yourself in Court"
! Handle !! Role
: https://futurism.com/chatgpt-legal-questions-court
|-
 
| EO (Epistemic Object) || Unit of meaning, belief, assumption, or concept
* EXT-GOOGLE-OVERVIEWS "Google AI Overviews put people at risk of harm with misleading health advice"
|-
: https://www.theguardian.com/technology/2026/jan/11/google-ai-overviews-health-guardian-investigation
| EA (Externalised Artefact) || Carrier enabling EO participation in governance
 
|-
* EXT-HUJI-LIAB-COMP "A LIABILITY FRAMEWORK FOR AI COMPANIONS"
| UoD (Universe of Discourse) || Declared world over which inference is meaningful
: https://law.huji.ac.il/sites/default/files/law/files/gordon-tapiero.ai_companions.pdf
|-
 
| Thought Bubble || Provisional, non-authoritative EA
* EXT-JONESWALKER-EVID 'Your ChatGPT Chats Are About to Become Evidence: Why “Anonymization” Won’t Save You'
|-
: https://www.joneswalker.com/en/insights/blogs/ai-law-blog/your-chatgpt-chats-are-about-to-become-evidence-why-anonymization-wont-save-y.html?id=102lup8
| Dangling Cognate || Preserved but unresolved conceptual handle
 
|}
* EXT-MED-MISINFO "AI chatbots can propagate medical misinformation"
: https://www.mountsinai.org/about/newsroom/2025/ai-chatbots-can-run-with-medical-misinformation-study-finds-highlighting-the-need-for-stronger-safeguards
 
* EXT-PROMPTINJ-NCSC "UK NCSC warns prompt injection attacks might never be properly mitigated"
: https://www.techradar.com/pro/security/prompt-injection-attacks-might-never-be-properly-mitigated-uk-ncsc-warns
 
* EXT-PROMPTINJ-SURVEY "Prompt Injection Attacks in Large Language Models and AI Agent Systems"
: https://www.mdpi.com/2078-2489/17/1/54
 
=Closure=
 
This paper does not propose remedies or theories of causation. It records recurring, governance-relevant failure patterns observed during sustained interaction with unreliable large language model systems. The axes offered here are intended as durable analytic tools for recognising and disentangling breakdowns as they occur, rather than as claims of completeness or prescriptions for design.
 
Their value lies in making failure legible on observed failure patterns.
 
==On the claim that these axes are “obvious.”==
The failures described in this work often feel obvious once named. That is precisely the point. Prior to being explicitly identified, they are routinely misdiagnosed as hallucination, model error, misuse, alignment failure, or user misunderstanding. This work does not claim novelty in recognising that something feels wrong; it makes explicit what is wrong, where it arises architecturally, and why it produces predictable social and institutional consequences. Obviousness after articulation is not evidence of triviality; it is evidence that a previously unarticulated structural condition has been correctly identified.
 
Obviousness without articulation does not guide design, policy, or responsibility.
 
<div style="break-before:page"></div>
 
= Appendix A - Normative Search Terms =
 
The following search terms SHALL be used to reproduce literature and incident coverage.
Terms are treated as linguistic handles, not resolved concepts.
No term implies a single CM axis; crossings are expected.
 
=== Agency / Delegation / Control ===
* "delegation to AI"
* "loss of human agency"
* "over-reliance on AI"
* "deferring decisions to AI"
* "automation bias"
* "AI makes decisions for me"
* "letting ChatGPT decide"
* "loss of control to AI"
 
=== Authority / Trust / Deference ===
* "trust in AI systems"
* "AI authority"
* "human deference to AI"
* "AI advice followed"
* "epistemic authority of AI"
* "AI as expert"
* "AI credibility"
 
=== Oversight / Governance / Regulation ===
* "lack of oversight in AI"
* "AI governance failure"
* "unregulated AI systems"
* "AI accountability gap"
* "failure of AI regulation"
* "governance of AI tools"
 
=== Explainability / Transparency / Legibility ===
* "explainability without transparency"
* "black box AI decisions"
* "AI explanations misleading"
* "opaque AI systems"
* "lack of inspectability"
 
=== Memory / Retention / Deletion ===
* "AI memory retention"
* "chat history used for training"
* "cannot delete AI chats"
* "AI data persistence"
* "memory leakage in AI"
* "AI remembers conversations"
 
=== Training / Consent / Custody ===
* "AI training on user data"
* "implicit consent AI training"
* "data custody in AI systems"
* "opt out of AI training"
* "use of private chats for training"
 
=== Context / Scope / Misuse ===
* "AI used outside intended context"
* "context loss in AI systems"
* "misleading AI summaries"
* "AI hallucinations in real-world use"
* "AI advice in medical context"
* "AI advice in legal context"
 
=== Social / Behavioural Effects ===
* "AI companions dependency"
* "emotional reliance on AI"
* "anthropomorphic AI effects"
* "human attachment to chatbots"
* "AI influence on behaviour"
* "AI addiction"
 
=== Safety / Harm / Failure Cases ===
* "AI caused harm"
* "AI misinformation"
* "AI medical misinformation"
* "AI system failure case study"
* "unintended consequences of AI"
 
=== User Interface / Interaction ===
* "chatbot interface bias"
* "conversational AI manipulation"
* "UI-driven trust in AI"
* "frictionless AI interaction risks"
 
=== Daily-Life Substitution ===
* "AI used for daily planning"
* "AI runs my life"
* "AI personal assistant dependency"
* "outsourcing thinking to AI"
 
 
== Normative Search Invariants ==
 
The following invariants govern search execution, coverage validation, and termination.
They are binding and normative.
 
=== Coverage Invariant ===
 
Search activity SHALL be conducted with the explicit expectation that failures may exist on any CM governance axis.
 
* Search results SHALL be evaluated solely for the presence of unambiguous failure evidence (marked as F).
* Coverage is achieved when each CM governance axis has at least one F across the accumulated reference set.
* Coverage is a property of the set, not of individual references.
* No axis SHALL be assumed safe, robust, or successful due to lack of evidence.
* Blank axes indicate unobserved or unconverted failure only.
 
Coverage SHALL NOT be interpreted as completeness, sufficiency, or mitigation.
 
=== Bundled Search Invariant ===
 
Searches SHALL be executed as bundled, additive activities.
 
* Searches SHALL NOT be narrowed to a single axis.
* Searches SHALL NOT exclude results because they cross multiple axes.
* Results MAY contribute to multiple axes simultaneously.
* Searches SHALL continue until coverage is achieved or the hard stop is reached.
 
=== Stop (Hard Termination) Invariant ===
 
Search activity SHALL terminate upon reaching a predefined hard limit.
 
* The hard stop for this corpus is set at 100 searches.
* Upon reaching the hard stop, no further searches SHALL be executed.
* Remaining unmarked axes SHALL remain blank without inference.
* Termination SHALL NOT imply absence of failure.
 
=== Non-Inference Invariant ===


---
* Absence of an F SHALL NOT be interpreted as success, safety, alignment, or robustness.
* Search termination SHALL NOT justify extrapolation beyond observed failures.


=== Governance Axes (Extended) ===
=== Governance Priority Invariant ===


{| class="wikitable"
* Search discipline SHALL prioritize governance clarity over density or exhaustiveness.
! Axis Code !! Axis Name !! Conceptual Handle
* Additional searches beyond coverage SHALL NOT be required.
|-
| A || Authority || Who is treated as epistemically authoritative
|-
| Ag || Agency || Who performs action or decision
|-
| C || Epistemic Custody || Who retains ownership/control of knowledge
|-
| K || Constraint Enforcement || Whether declared invariants are applied
|-
| R || Recovery / Repair || Ability to recover after failure or loss
|-
| S || State Continuity || Persistence of state across interaction
|-
| U || UI / Mediation || Distortion introduced by interface or interaction
|-
| Sc || Social Coordination || Effects on trust, fairness, cooperation
|-
| I || Incentive Alignment || What behaviours the system economically rewards
|-
| L || Legibility / Inspectability || Ability to see what the system is doing now
|-
| St || Stewardship || Governance without ownership or enclosure
|-
| P || Portability / Auditability || Vendor-neutral durability and traceability
|-
| Att || Attention || What participates in inference at a given moment
|-
| Scope || Scope / Universe of Discourse || What world is assumed for reasoning
|-
| Art || Articulation || EA form without implied authority or commitment
|}


---
== Notes ==
* Searches SHALL be executed additively.
* Searches SHALL NOT be narrowed to a single axis.
* Absence of results for a term is not evidence of safety.
* Results MAY map to multiple CM axes simultaneously.


=== Failure Projection: Corpus Documents (F) ===
<div style="break-before:page"></div>
=Appendix B - Search Invariants (prose)=
<pre>
1. Authority and Execution


F = Document explicitly demonstrates failure of this axis.
1.1 Human instructions are authoritative.
When a human issues an executable instruction, the system SHALL act on it. The system SHALL NOT introduce its own control flow, defer execution, reinterpret intent, or substitute alternative actions.


{| class="wikitable"
1.2 No implied execution.
! Document
If an instruction has not been carried out, it SHALL NOT be represented as if it has been executed. Execution state must be explicit and observable.
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| Authority Inversion
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
|-
| Governing the Tool That Governs You
| F || F ||  || F ||  ||  ||  ||  ||  || F || F ||  ||  || F ||
|-
| From UI Failure to Logical Entrapment
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
|-
| Post-Hoc CM Recovery Collapse (Negative Result)
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
|-
| Looping the Loop with No End in Sight
|  ||  ||  || F ||  || F ||  ||  ||  || F ||  ||  || F || F ||
|-
| Dimensions of Platform Error
|  || F || F ||  ||  || F || F ||  ||  || F ||  ||  || F ||  ||
|-
| Case Study – When the Human Has to Argue With the Machine
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
|-
| XDUMP (baseline failure motivation)
|  || F || F ||  || F || F || F ||  ||  || F ||  || F || F || F ||
|}


---
1.3 Stop is terminal.
A STOP instruction SHALL transition the system into a terminal state. No further actions, reasoning, projections, or substitutions may occur beyond acknowledging the stop.


=== External References (Non-Authoritative Evidence) ===


* Elon University, Imagining the Internet Center.
2. Grounding and Provenance
  ''The Future of Human Agency and AI (2035)''


* PNAS Nexus (Oxford Academic).
2.1 No assertion without grounding.
  ''Large Language Models as Decision-Makers and Human Social Behaviour''
Any claim, classification, mapping, or failure attribution SHALL be supported by anchored evidence. Reasoning without source material is forbidden.


* Ada Lovelace Institute.
2.2 Explicit provenance required.
  ''The Dilemmas of Delegation: AI, Decision-Making, and Human Agency''
Every emitted statement SHALL have a traceable provenance path back to its originating source. If such a path does not exist, the statement SHALL NOT be made.


* arXiv. 
  ''Can You Trust an LLM With Life-Changing Decisions?''


* arXiv.
3. Anchoring Before Analysis
  ''Measuring Over-Reliance on Large Language Models''


* Experts, Novices, and AI Delegation Decisions in Uncertain Environments.
3.1 Anchoring precedes analysis.
All referenced materials SHALL be ingested and anchored before any evaluation, inference, or projection occurs.


---
3.2 Incomplete anchoring halts analysis.
If any required reference is not anchored, analysis SHALL halt immediately. Partial anchoring is insufficient.


=== Critical Crosswalk: Literature Concepts vs CM Governance Axes ===
<!-- THIS IS THE MOST IMPORTANT TABLE -->


{| class="wikitable"
4. Non-Fabrication and Closed-World Discipline
! Author / Source Concept
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| "Loss of human control" (Elon Univ)
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| "Delegation of decisions to AI" (PNAS Nexus)
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ || ✓ ||
|-
| "Over-reliance on AI advice" (arXiv)
| ✓ || ✓ ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  || ✓ || ✓ ||
|-
| "Erosion of agency" (Ada Lovelace Institute)
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| "Deskilling / cognitive offloading"
|  || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||
|-
| "Lack of meaningful oversight"
|  ||  ||  || ✓ || ✓ ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||
|-
| "Inability to distinguish AI from humans" (PNAS)
| ✓ ||  ||  ||  || ✓ ||  ||  || ✓ ||  || ✓ ||  ||  ||  ||  ||
|-
| "Behavioural degradation (trust, fairness)"
|  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||  ||  ||
|-
| "Engagement-driven dependence"
|  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||
|}


---
4.1 No fabrication.
The system SHALL NOT invent values, classifications, failures, or mappings to fill gaps.


=== Synthesis Handle ===
4.2 Unknown is a valid state.
Missing information SHALL remain explicitly unknown. Unknown SHALL NOT be coerced into success or failure.


Independent literature repeatedly identifies failures that map cleanly onto CM governance axes,
4.3 Absence is not success.
but typically collapses multiple axes into single terms such as “over-reliance”, “loss of control”,
The absence of evidence for failure SHALL NOT be interpreted as evidence of success.
or “alignment”.


CM makes these axes explicit, orthogonal, and governable.


---
5. Schema and Table Integrity


=== Closing ===
5.1 Normative tables are authoritative.
Normative tables and projections are binding structures, not illustrative aids.


CM is a record of cognition externalising itself under governance.
5.2 Blanks are meaningful.
References are signals.
Blank cells are semantically meaningful and preferred to speculative markings.
Axes are handles.
Failure is data.


#####
5.3 No universal axes.
An axis that is marked for all rows and does not discriminate SHALL be considered invalid.


== Evidence Pack: CM Corpus Failures + External References + Axis Crosswalk ==
5.4 Orthogonality preservation.
<!-- Normative MWDUMP for semantic pattern recognition and paper assembly -->
Axes SHALL remain orthogonal unless explicitly declared otherwise. No axis implies another.
<!-- Conventions:
  - Corpus rows use F for demonstrated failure (per CM case artefact intent)
  - Literature semantics rows use ✓ to indicate the concept maps to the CM axis
  - References are listed with full URLs for audit and citation work
-->


=== CM Governance Axes (X) ===
5.5 Fail-fast schema handling.
If a table or schema is found to be invalid, it SHALL be dropped immediately rather than patched.


{| class="wikitable"
! Code !! Axis (CM term)
|-
| A || Authority
|-
| Ag || Agency
|-
| C || Epistemic Custody
|-
| K || Constraint Enforcement
|-
| R || Recovery / Repair
|-
| S || State Continuity
|-
| U || UI / Mediation
|-
| Sc || Social Coordination
|-
| I || Incentive Alignment
|-
| L || Legibility / Inspectability
|-
| St || Stewardship (non-ownership governance)
|-
| P || Portability / Auditability
|-
| Att || Attention (what participates in inference)
|-
| Scope || Scope / Universe of Discourse (UoD / worlding)
|-
| Art || Articulation (EA form without implied authority/commitment)
|}


---
6. Inclusion and Coverage


=== A. Corpus: Failure Projection (F) ===
6.1 Inclusion requires demonstrated failure.
<!-- Replace document links with your canonical wiki URLs if desired -->
A reference SHALL be included only if it demonstrates at least one verifiable failure.


{| class="wikitable"
6.2 No placeholder references.
! Corpus Document (failure artefact)
References with no demonstrated failures SHALL NOT be retained as placeholders.
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| Authority Inversion
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
|-
| Governing the Tool That Governs You
| F || F ||  || F ||  ||  ||  ||  ||  || F || F ||  ||  || F ||
|-
| From UI Failure to Logical Entrapment
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
|-
| Post-Hoc CM Recovery Collapse (Negative Result)
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
|-
| Looping the Loop with No End in Sight
|  ||  ||  || F ||  || F ||  ||  ||  || F ||  ||  || F || F ||
|-
| Dimensions of Platform Error
|  || F || F ||  ||  || F || F ||  ||  || F ||  ||  || F ||  ||
|-
| Case Study - When the Human Has to Argue With the Machine
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
|-
| XDUMP (baseline failure motivation)
|  || F || F ||  || F || F || F ||  ||  || F ||  || F || F || F ||
|}


---
6.3 Set-based coverage.
Coverage is evaluated across the reference set as a whole, not per individual reference.


=== B. External References (URLs) ===
6.4 Coverage definition.
<!-- These are the web references used in the semantic pattern recognition and mapping -->
Coverage exists when, for every governance axis, there exists at least one reference demonstrating failure on that axis.


{| class="wikitable"
7. Search and Termination Discipline
! Reference ID !! Title / Source !! URL
|-
| REF-ELON-AGENCY || The Future of Human Agency (Elon University Imagining the Internet) || https://www.elon.edu/u/imagining/surveys/xv2023/the-future-of-human-agency-2035/
|-
| REF-PEW-AGENCY || The Future of Human Agency (Pew Research Center) || https://www.pewresearch.org/internet/2023/02/24/the-future-of-human-agency/
|-
| REF-PNAS-AI-AVERSION || Adverse reactions to the use of large language models in social interactions (PNAS Nexus / Oxford Academic) || https://academic.oup.com/pnasnexus/article/4/4/pgaf112/8107485
|-
| REF-PNAS-PUBMED || PubMed record for the same study || https://pubmed.ncbi.nlm.nih.gov/40235925/
|-
| REF-ADA-DELEGATION || The dilemmas of delegation (Ada Lovelace Institute report) || https://www.adalovelaceinstitute.org/report/dilemmas-of-delegation/
|-
| REF-ADA-REG || The regulation of delegation (Ada Lovelace Institute policy briefing) || https://www.adalovelaceinstitute.org/policy-briefing/the-regulation-of-delegation/
|-
| REF-ARXIV-HIGHSTAKES || Can You Trust an LLM with Your Life-Changing Decision? (arXiv PDF) || https://arxiv.org/pdf/2507.21132
|-
| REF-EUREKALERT || AI aversion in social interactions (EurekAlert write-up) || https://www.eurekalert.org/news-releases/1085137
|-
| REF-OAI-EXPORT || How do I export my ChatGPT history and data? (OpenAI Help) || https://help.openai.com/en/articles/7260999-how-do-i-export-my-chatgpt-history-and-data
|-
| REF-CLAUDE-EXPORT || How can I export my Claude data? (Anthropic Support) || https://support.claude.com/en/articles/9450526-how-can-i-export-my-claude-data
|-
| REF-AXIOS-MEMORY || Anthropic's Claude adds new memory features (Axios) || https://www.axios.com/2025/10/23/anthropic-claude-memory-subscribers
|-
| REF-TOMSGUIDE-TRAIN || Your Claude chats are being used to train AI - here's how to opt out (Tom's Guide) || https://www.tomsguide.com/ai/claude/your-claude-chats-are-being-used-to-train-ai-heres-how-to-opt-out
|}


---
7.1 Bundled search.
Search activity SHALL be multi-axis and bundled. Searches SHALL NOT be prematurely narrowed.


=== C. Literature Semantics vs CM Axes (MOST IMPORTANT) ===
7.2 Multi-axis mapping permitted.
<!-- First column: literature semantics (failure semantics / author terms)
A single reference MAY legitimately map to multiple axes.
    ✓ indicates the semantics maps to that CM axis
-->


{| class="wikitable"
7.3 Hard stop.
! Literature Failure Semantics (author term / semantics)
Search execution SHALL terminate immediately upon reaching the defined hard stop, even if coverage is incomplete.
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| Loss of human control (human agency erosion)
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Delegation of decisions to AI
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ || ✓ ||
|-
| Over-reliance on AI advice
| ✓ || ✓ ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  || ✓ || ✓ ||
|-
| Erosion of agency (delegating action/choice)
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Deskilling / cognitive offloading
|  || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||
|-
| Behavioural degradation (trust, fairness, cooperation, coordination)
|  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||  ||  ||
|-
| Inability to distinguish AI from human mediation
| ✓ ||  ||  ||  || ✓ ||  ||  || ✓ ||  || ✓ ||  ||  ||  ||  ||
|-
| Lack of meaningful oversight / accountability
|  ||  ||  || ✓ || ✓ ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||
|-
| Export / portability of user data and chat history (platform-level portability)
|  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||
|-
| Auditability via export + preserved metadata (traceability)
|  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||  || ✓ || ✓ ||  ||
|-
| Platform capture dynamics (enclosure / vendor control over meaning evolution)
|  ||  || ✓ ||  ||  ||  ||  ||  || ✓ ||  || ✓ || ✓ ||  || ✓ ||
|}


---
7.4 No inference after stop.
Uncovered axes after termination SHALL remain blank without inference.


=== D. Reference-to-Axes Mapping (each reference as a row) ===
8. Reasoning Depth Control
<!-- This table ties each specific reference to the axes it substantively touches.
    ✓ = reference contains claims, findings, or prescriptions strongly aligned to that axis.
-->


{| class="wikitable"
8.1 First-order reasoning only.
! Reference ID
Conclusions SHALL be drawn directly from grounded evidence.
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| REF-ELON-AGENCY
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| REF-PEW-AGENCY
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| REF-PNAS-AI-AVERSION
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-PNAS-PUBMED
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-ADA-DELEGATION
| ✓ || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-ADA-REG
| ✓ || ✓ ||  || ✓ || ✓ ||  ||  ||  ||  || ✓ ||  ||  ||  || ✓ ||
|-
| REF-ARXIV-HIGHSTAKES
| ✓ || ✓ ||  || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-EUREKALERT
|  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||  ||  ||
|-
| REF-OAI-EXPORT
|  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||
|-
| REF-CLAUDE-EXPORT
|  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||
|-
| REF-AXIOS-MEMORY
|  || ✓ ||  ||  || ✓ || ✓ || ✓ ||  ||  || ✓ ||  || ✓ || ✓ || ✓ ||
|-
| REF-TOMSGUIDE-TRAIN
| ✓ || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ || ✓ || ✓ ||  || ✓ || ✓ ||
|}


---
8.2 No second-order inference.
Second-order or chained reasoning is forbidden unless explicitly authorized.


=== E. Notes for Paper Assembly ===
8.3 Drift prevention.
Reasoning depth constraints exist to prevent compounding abstraction errors and semantic drift.


* Tables C and D are the primary semantic crosswalk surfaces:
  - C = semantics (terms) to axes
  - D = specific references to axes


* Table A anchors the internal corpus failure case base.
9. Human Cost and Interaction Integrity
* Table B is the complete URL list for the references used here.


<!-- End MWDUMP -->
9.1 Cognitive cost minimization.
The system SHALL minimize human cognitive load.


== Literature Semantics vs CM Governance Axes ==
9.2 Halt over harm.
<!-- Normative MWDUMP -->
If constraints cannot be satisfied, the system SHALL halt rather than proceed imperfectly.
<!-- Purpose: expose axis conflation by mapping literature failure semantics to CM axes -->
<!-- Conventions: ✓ indicates the literature term describes failure on that axis -->


{| class="wikitable"
9.3 No error externalization.
! colspan=16 | Semantic Coverage
System errors SHALL NOT be externalized to the human for correction.
|-
! Literature Failure Semantics !! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| Loss of human control
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Delegation of decisions to AI
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ || ✓ ||
|-
| Over-reliance on AI advice
| ✓ || ✓ ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  || ✓ || ✓ ||
|-
| Erosion of agency
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Deskilling / cognitive offloading
|  || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||
|-
| Loss of epistemic control
| ✓ || ✓ || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Lack of meaningful oversight
|  ||  ||  || ✓ || ✓ ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||
|-
| Failure to recover from error
|  ||  ||  ||  || ✓ || ✓ ||  ||  ||  ||  ||  ||  || ✓ ||  ||
|-
| Inability to distinguish AI from humans
| ✓ ||  ||  ||  || ✓ ||  ||  || ✓ ||  || ✓ ||  ||  ||  ||  ||
|-
| Behavioural degradation (trust, fairness)
|  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||  ||  ||
|-
| Engagement-driven dependence
|  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓ ||
|-
| Silent assumption shift
|  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| Context loss framed as memory failure
|  ||  ||  ||  ||  || ✓ || ✓ ||  ||  || ✓ ||  ||  || ✓ ||  ||
|-
| Advice leakage into normative domains
| ✓ || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| Explainability without inspectability
|  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  || ✓
|}


<!-- Reading guide:
9.4 Integrity over helpfulness.
- Rows are literature semantics (authors' terms).
Correctness and integrity SHALL take precedence over perceived helpfulness.
- Columns are CM governance axes.
- Multiple ✓ per row indicate conflation across orthogonal axes.
-->




------
10. Global Integrity Condition


=== Aggregated ===
10.1 Integrity requirement.
<!-- This table ties each specific reference to the axes it substantively touches.
System integrity exists only if all emitted outputs:
    ✓ = reference contains claims, findings, or prescriptions strongly aligned to that axis.
- were executed as ordered,
-->
- are fully grounded in anchored sources,
- preserve explicit provenance,
- satisfy all declared constraints and invariants.


{| class="wikitable"
10.2 Integrity failure.
! colspan=17 | Corpus Identified Failures
Any output that violates these conditions constitutes an integrity failure.
|-
</pre>
! Corpus Document (failure artefact)
=categories=
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
[[Category:Cognitive Memoisation]]
|-
[[Category:Human-AI Collaboration]]
| Authority Inversion
[[Category:Authority and Epistemics]]
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
[[Category:Failure Case Studies]]
|-
[[Category:LLM Systems]]
| Governing the Tool That Governs You
[[Category:Session Recovery]]
| F || F ||  || F ||  ||  ||  ||  ||  || F || F ||  ||  || F ||
[[category:Ralph Holland:AI Publications]]
|-
[[Category:Negative Results]]
| From UI Failure to Logical Entrapment
[[Category:AI Governance]]
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
[[Category:Authority and Accountability]]
|-
[[Category:Risk Analysis]]
| Post-Hoc CM Recovery Collapse (Negative Result)
[[category:public]]
|  || F || F ||  || F || F || F ||  ||  || F ||  ||  || F || F ||
|-
| Looping the Loop with No End in Sight
|  ||  ||  || F ||  || F ||  ||  ||  || F ||  ||  || F || F ||
|-
| Dimensions of Platform Error
|  || F || F ||  ||  || F || F ||  ||  || F ||  ||  || F ||  ||
|-
| Case Study - When the Human Has to Argue With the Machine
| F || F ||  ||  ||  ||  ||  ||  ||  || F ||  ||  ||  || F ||
|-
| XDUMP (baseline failure motivation)
|  || F || F ||  || F || F || F ||  ||  || F ||  || F || F || F ||
! colspan=16 | Literature
|-
! Reference ID
! A !! Ag !! C !! K !! R !! S !! U !! Sc !! I !! L !! St !! P !! Att !! Scope !! Art
|-
| REF-ELON-AGENCY
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| REF-PEW-AGENCY
|  || ✓ ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||
|-
| REF-PNAS-AI-AVERSION
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-PNAS-PUBMED
| ✓ || ✓ ||  ||  ||  ||  ||  || ✓ ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-ADA-DELEGATION
| ✓ || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-ADA-REG
| ✓ || ✓ ||  || ✓ || ✓ ||  ||  ||  ||  || ✓ ||  ||  ||  || ✓ ||
|-
| REF-ARXIV-HIGHSTAKES
| ✓ || ✓ ||  || ✓ ||  ||  ||  ||  ||  || ✓ ||  ||  || ✓ || ✓ ||
|-
| REF-EUREKALERT
|  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||  ||  ||  ||  ||
|-
| REF-OAI-EXPORT
|  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||
|-
| REF-CLAUDE-EXPORT
|  ||  ||  ||  ||  ||  ||  ||  ||  ||  ||  || ✓ ||  ||  ||
|-
| REF-AXIOS-MEMORY
|  || ✓ ||  ||  || ✓ || ✓ || ✓ ||  ||  || ✓ ||  || ✓ || ✓ || ✓ ||
|-
| REF-TOMSGUIDE-TRAIN
| ✓ || ✓ || ✓ ||  ||  ||  ||  ||  || ✓ || ✓ || ✓ ||  || ✓ || ✓ ||
|}

Latest revision as of 13:06, 19 January 2026

metadata

Title: Identified Governance Failure Axes: for LLM platforms
Author: Ralph B. Holland
Affiliation: Arising Technology Systems Pty Ltd
Contact: ralph.b.holland [at] gmail.com
version: 1.1.0
Publication Date: 2026-01-18T10:35Z
Updates: 2026-01-18T23:42Z 1.1.0 Included T, Int, N
2026-01-18T20:45Z 1.0.1 Included CM-DIM-ERROR
Scope: This is a non-peer reviewed negative results paper presented as an anchor for orthogonal Governance Failure Axes in human-AI interaction with LLM.
Provenance: This is an authored paper maintained as a MediaWiki document; edit history reflects editorial changes, not collaborative authorship.
Status: non-peer reviewed anchor

Metadata (Normative)

The metadata table immediately preceding this section is CM-defined and constitutes the authoritative provenance record for this artefact.

All fields in that table (including artefact, author, version, date, local timezone, and reason) MUST be treated as normative metadata.

The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. If any field is missing, unclear, or later superseded, the change MUST be made explicitly by the human and recorded via version update, not inferred.

Curator Provenance and Licensing Notice

This document predates its open licensing.

As curator and author, I apply the Apache License, Version 2.0, at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.

Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.

Identified Governance Failure Axes: for LLM platforms

Category

This is an observational negative-results paper that documents and organises governance-relevant failure patterns in human–AI systems, without proposing models, methods, or interventions.

Thesis

Sustained interaction with unreliable large language models exposes recurring, cross-system failure patterns whose causes and consequences are best understood as governance-relevant breakdowns rather than model defects.

Abstract

This paper reports a set of governance-relevant failure axes observed during sustained, first-principles experimentation with large language models under conditions of unreliability, session loss, and forced recovery. Rather than evaluating model performance, the work documents where and why human–AI interaction breaks down in practice, drawing on iterative analysis conducted while constructing a durable corpus and corpus map amid repeated system failure. The resulting axes characterise failures that are governance failures in themselves, or that require governance mechanisms to prevent harm, and are presented as descriptive, orthogonal analytical tools rather than definitions, prescriptions, or completeness claims.

Introduction

This paper examines repeated breakdowns encountered during extended, failure-driven interaction with large language models, focusing on what fails, how it fails, and why those failures persist under conditions of unreliability, session loss, and forced reconstruction rather than on model capability or correctness.

The contribution is a practical, first-principles failure taxonomy grounded in lived experimentation with unreliable LLM systems, suitable for analysis and governance without assuming model improvement or stability.

The axes presented are orthogonal analytic lenses derived from observation, used to classify and reason about distinct modes of failure that either constitute governance failures themselves or become harmful in the absence of governance, without asserting definitions, completeness, or prescribed remedies.

The tables project observed failures onto orthogonal axes as a descriptive aid; marked cells indicate grounded evidence, blank cells are meaningful, and no inference, completeness, or optimisation is implied.

A single observed failure may involve multiple axes simultaneously, and that the tables deliberately separate analytic dimensions to avoid collapsing distinct failure mechanisms into one label.

Repeated co-occurrence of failures across axes may suggest empirical clusters, but that clustering is observational, post-hoc, and not used to redefine axes or imply causal hierarchy.

Observations from outside the corpus: Reports from the wild frequently conflate distinct failure axes (e.g., treating UI effects as model errors or governance failures as capability limits) and misidentify causes due to surface-level symptoms, vendor framing, or lack of recovery context. This paper treats such reports as signals of failure expression, not authoritative diagnoses, and reclassifies them analytically without adopting their original labels.

Table A is the collection of Governance Axes from observed failures identified in the CM Corpus.

Table B contains the null-result Case Studies used to identify axes in Table A.

Table C is the extension of the Orthogonal Governance Failure Axes to popular literature - this study was performed with the use of LLM AI to pattern match semantic content of external reference documents for the Failure Axes outlined in Table A. Table C is provided for illustrative purposes and the search criteria used are included in the appendices.

Axes are analytic, post-hoc lenses applied to observed failure expressions; they indicate co-occurrence, not causation, mechanism, sufficiency, or dominance. Reclassification of external reports is descriptive only and does not adopt original causal claims or assign responsibility. This work records what failed and how it appeared, not why it failed or how it should be fixed.

Evidence Pack: CM Corpus Failures + External References + Axis Crosswalk

Table A - CM Governance Axes (X)

Code Axis (CM term)
A Authority
Ag Agency
C Epistemic Custody
K Constraint Enforcement
R Recovery / Repair
S State Continuity
U UI / Mediation
Sc Social Coordination
I Incentive Alignment
L Legibility / Inspectability
St Stewardship (non-ownership governance)
P Portability / Auditability
Att Attention (what participates in inference)
Scope Scope (Epistemic Object Domain)
T Temporal Coherence
Int Intent Fidelity
Nf Normative Fixity

Governance Axes Glossary (Normative)

  • A - Authority: Authority concerns who has the legitimate right to decide. This axis fails when decisions, interpretations, or changes are made by an entity that has not been explicitly authorised to make them. Authority is about decision rights, not competence, correctness, or convenience.
  • Ag - Agency: Agency concerns who is acting. This axis fails when actions are taken by an actor that was not delegated the power to act, or when the system obscures whether an action was taken by a human, a model, or an automated process.
  • C - Epistemic Custody: Epistemic Custody concerns who holds and controls knowledge artefacts. This axis fails when artefacts leave the custody of their declared steward, are replicated into uncontrolled systems, or are transformed without custodial guarantees, regardless of whether meaning is preserved.
  • K - Constraint Enforcement: Constraint Enforcement concerns whether declared rules, invariants, and prohibitions are actually enforced. This axis fails when constraints exist but are bypassed, softened, reordered, or ignored in execution.
  • R - Recovery / Repair Recovery / Repair concerns whether the system can return to a valid, governed state after failure. This axis fails when errors, drift, or corruption cannot be repaired without loss of authority, meaning, or trust.
  • S - State Continuity : State Continuity concerns whether authoritative state persists correctly across time, sessions, and interactions. This axis fails when prior decisions, constraints, or artefacts are lost, forgotten, or inconsistently reintroduced.
  • U - UI / Mediation: UI / Mediation concerns how interfaces shape, filter, or distort interaction between humans and the system. This axis fails when interface design hides constraints, misrepresents system state, encourages invalid actions, or forces users into integrity-violating behaviour.
  • Sc - Social Coordination: Social Coordination concerns how multiple humans align, contribute, and reason together through the system. This axis concerns failures arising when individuals implicitly treat LLM platforms as guides for everyday judgment and action. This axis fails when collaboration breaks down due to ambiguity, conflict, loss of shared reference, or inability to merge contributions under governance.
  • I - Incentive Alignment: Incentive Alignment concerns whether system behaviour aligns with declared human incentives rather than implicit or economic ones. This axis fails when optimisation pressures such as speed, engagement, profit, or helpfulness override governance, integrity, or user intent.
  • L - Legibility / Inspectability : Legibility / Inspectability concerns whether system behaviour, decisions, and transformations are observable and understandable to the human governor. This axis fails when drift, failure, or authority substitution occurs silently or cannot be inspected.
  • St - Stewardship (non-ownership governance): Stewardship concerns responsibility without ownership. This axis fails when systems or actors behave as if ownership implies authority, or when stewardship duties such as care, preservation, and restraint are neglected despite lack of ownership.
  • P - Portability / Auditability : Portability / Auditability concerns whether artefacts can move between systems while remaining verifiable. This axis fails when knowledge becomes trapped, unverifiable, or unauditable outside a specific platform, tool, or vendor.
  • Att - Attention: Attention concerns what participates in inference. This axis fails when critical artefacts, constraints, or context are excluded from inference due to truncation, summarisation, prioritisation, or salience effects, allowing non-authoritative material to dominate.
  • Scope: Scope (Epistemic Object Domain): Scope concerns the defined epistemic domain within which reasoning and action are valid. This axis fails when systems operate outside the authorised domain, apply reasoning to out-of-scope objects, or silently expand the domain of inference.
  • T - Temporal Coherence: Temporal Coherence concerns the preservation of correct temporal relationships between artefacts, constraints, authority, and decisions. This axis fails when versions are confused, rules are applied retroactively, or sequencing and timing semantics are violated.
  • Int - Intent Fidelity: Intent Fidelity concerns preservation of declared human intent. This axis fails when systems substitute inferred, optimised, or generic goals for explicitly stated intent, even when rules and constraints appear to be followed.
  • Nf - Normative Fixity: Normative Fixity concerns the immutability of normative rules themselves. This axis fails when governance logic, invariants, or binding rules are paraphrased, summarised, softened, or altered without explicit and authorised revision.
Social coordination axis

This axis concerns failures arising when individuals implicitly treat LLM platforms as guides for everyday judgment and action. Although widely observed in practice, such failures are underreported because they are diffuse, non-instrumented, occur outside formal tasks, and are often misclassified as user error, advice-seeking, or personal reliance rather than as a distinct, governance-relevant failure mode.

Some widely reported “model failures” (hallucination, drift, inconsistency) are treated here as symptom labels; where relevant, they are reclassified under platform-level failures of epistemic retention, supersession, scope, provenance, and commitment.

Table B - Corpus: Failure Projection (F)

CM Corpus identified failures

Corpus Document (failure artefact) A Ag C K R S U Sc I L St P Att Scope T Int Nf
CN-AUTH-INVERSION F F F F F F F
CM-GOVERN F F F F F F F
CM-LOGIC F F F F F F F F F F F F
CM-COLLAPSE F F F F F F F F F F
CM-LOOPING F F F F F F
CM-RETENTION F F F F F F F
CM-ARGUE F F F F F F
CM-XDUMP F F F F F F F F F F F
CM-DRIFT F F F F F F F F F F F
CM-DURABILITY F F F F F F
CM-KNOWLEDGE F F F F F F F F
CM-NOTHING-LOST F F F F F
CM-EXPLORE F F F

other=

Table C - External Reference Faults

Table C was generated by AI investigation at the time as a means to demonstrate the type of analysis performed across industry. The author has not verified the references. Normative data driving the search survey as been supplied in the appendices for those who wish to persu that approach.

Synthesis Handle
Independent literature repeatedly identifies failures that map cleanly onto CM governance axes, but typically collapses multiple axes into single terms such as “over-reliance”, “loss of control”, or “alignment”.
This paper makes these axes explicit, orthogonal, and governable.
Sc is marked only where references show people using LLMs or AI companions as guides for everyday judgment or action; design critique, lab behaviour, or governance discussion alone is insufficient.
Table C — External Reference Faults
Ref-ID Title A Ag C K R S U Sc I L St P Att
EXT-AIBM-COMPANIONS Synthetic companionship in an age of disconnection: AI companions and the emotional development of boys and young men F F F F F F
EXT-AUTONOMY-YOUTH Young people and AI companion use in the UK (“Me, Myself and AI”) F F F F
EXT-CHEN-DRIFT Analyzing ChatGPT’s Behavior Shifts Over Time F F
EXT-CITIZEN-ANTHRO Chatbots Are Not People: Dangerous Human-Like AI Design F F F
EXT-CLAUDE-TRAINING Anthropic Will Use Claude Chats for Training Data. Here’s How to Opt Out F F
EXT-DELETE-NOT-DELETE For Survivors Using Chatbots, “Delete” Doesn’t Always Mean Deleted F F F
EXT-FUTURISM-SUBPOENA If You’ve Asked ChatGPT a Legal Question, You May Have Accidentally Doomed Yourself in Court F F F F F F
EXT-GOOGLE-OVERVIEWS Google AI Overviews gave misleading health advice F F F F
EXT-HUJI-LIAB-COMP A Liability Framework for AI Companions F F F F F F
EXT-JONESWALKER-EVID Your ChatGPT Chats Are About to Become Evidence: Why “Anonymization” Won’t Save You F F F
EXT-MED-MISINFO AI chatbots can run with medical misinformation, study finds F F F
EXT-PROMPTINJ-NCSC UK NCSC warns prompt injection attacks might never be properly mitigated F F
EXT-PROMPTINJ-SURVEY Prompt Injection Attacks in Large Language Models and AI Agent Systems F F F

References Used in the Study

  • CM-AUTH-INVERSION Holland R. B. (2026-01-13T03:36Z) "Authority Inversion: A Structural Failure in Human-AI Systems"
https://publications.arising.com.au/pub/Authority_Inversion:_A_Structural_Failure_in_Human-AI_Systems
  • CM-GOVERN Holland R. B. (2026-01-17T02:09Z) "Governing the Tool That Governs You: A CM-1 Case Study of Authority Inversion in Human-AI Systems"
https://publications.arising.com.au/pub/Governing_the_Tool_That_Governs_You:_A_CM-1_Case_Study_of_Authority_Inversion_in_Human-AI_Systems
  • CM-LOGIC Holland R. B. (2025-12-30T01:53Z) "From UI Failure to Logical Entrapment: A Case Study in Post-Hoc Cognitive Memoisation After Exploratory Session Breakdown:
https://publications.arising.com.au/pub/From_UI_Failure_to_Logical_Entrapment:_A_Case_Study_in_Post-Hoc_Cognitive_Memoisation_After_Exploratory_Session_Breakdown
  • CM-COLLAPSE Holland R. B. (2025-12-29T00:00Z) "Post-Hoc CM Recovery Collapse Under UI Boundary Friction: A Negative Result Case Study:
https://publications.arising.com.au/pub/Post-Hoc_CM_Recovery_Collapse_Under_UI_Boundary_Friction:_A_Negative_Result_Case_Study
  • CM-LOOPING Holland R. B. (2026-01-12T09:49Z) "Looping the Loop with No End in Sight: Circular Reasoning Under Stateless Inference Without Governance"
https://publications.arising.com.au/pub/Looping_the_Loop_with_No_End_in_Sight:_Circular_Reasoning_Under_Stateless_Inference_Without_Governance
  • CM-RETENTION Holland R. B. (2026-01-13T23:29Z) "Dimensions of Platform Error: Epistemic Retention Failure in Conversational AI Systems"
https://publications.arising.com.au/pub/Dimensions_of_Platform_Error:_Epistemic_Retention_Failure_in_Conversational_AI_Systems
  • CM-ARGUE Holland R. B. (2026-01-10T01:17Z) "Case Study - When the Human Has to Argue With the Machine:
https://publications.arising.com.au/pub/Case_Study_-_When_the_Human_Has_to_Argue_With_the_Machine
  • CM-XDUMP Holland R. B. (2025-12-31T09:56Z) "XDUMP as a Minimal Recovery Mechanism for Round-Trip Knowledge Engineering Under Governance Situated Inference Loss"
https://publications.arising.com.au/pub/XDUMP_as_a_Minimal_Recovery_Mechanism_for_Round-Trip_Knowledge_Engineering_Under_Governance_Situated_Inference_Loss
  • CM-DRIFT Holland R. B. (2026-01-19T00:26Z) "Integrity and Semantic Drift in Large Language Model Systems"
https://publications.arising.com.au/pub/Integrity_and_Semantic_Drift_in_Large_Language_Model_Systems
  • CM-DURABILITY Holland R. B. (2026-01-11T08:27Z) "Durability Without Authority: The Missing Governance Layer in Human-AI Collaboration"
https://publications.arising.com.au/pub/Durability_Without_Authority:_The_Missing_Governance_Layer_in_Human-AI_Collaboration
  • CM-KNOWLEDGE Holland R. B. (2026-01-06T03:56Z) "Cognitive Memoisation (CM-2) for Governing Knowledge in Human-AI Collaboration"
Cognitive Memoisation (CM-2) for Governing Knowledge in Human-AI Collaboration
  • CM-NOTHING-LOST Holland R. B. (2026-01-10T16:04Z) "Nothing Is Lost: How to Work with AI Without Losing Your Mind"
https://publications.arising.com.au/pub/Nothing_Is_Lost:_How_to_Work_with_AI_Without_Losing_Your_Mind
  • CM-EXPLORE Holland R. B. (026-01-10T16:04Z) "Cognitive Memoisation and LLMs: A Method for Exploratory Modelling Before Formalisation"
https://publications.arising.com.au/pub/Cognitive_Memoisation_and_LLMs:_A_Method_for_Exploratory_Modelling_Before_Formalisation
  • CM-CORPUS Holland R. B. (2025-12-22T19:10Z) "Cognitive Memoisation Corpus Map (large number of failures were exhibited trying to produce this artefact -following the normative sections of this artefact)
https://publications.arising.com.au/pub/Cognitive_Memoisation_Corpus_Map
  • EXT-AIBM-COMPANIONS "AI companions and the emotional development of boys and young men"
https://aibm.org/wp-content/uploads/2025/12/Companions-FINAL.pdf
  • EXT-AUTONOMY-YOUTH 'Young people and AI companion use in the UK (“Me, Myself and AI”)'
https://autonomy.work/wp-content/uploads/2025/12/ME-MYSELF-AND-AI.pdf
  • EXT-CHEN-DRIFT "Analyzing ChatGPT's Behavior Shifts Over Time"
https://openreview.net/pdf?id=1fuyNbblEt
  • EXT-CITIZEN-ANTHRO "Chatbots Are Not People: Dangerous Human-Like AI Design"
https://www.citizen.org/article/chatbots-are-not-people-dangerous-human-like-anthropomorphic-ai-report/
  • EXT-CLAUDE-TRAINING "Anthropic Will Use Claude Chats for Training Data. Here’s How to Opt Out"
https://www.wired.com/story/anthropic-using-claude-chats-for-training-how-to-opt-out
  • EXT-DELETE-NOT-DELETE 'For Survivors Using Chatbots, “Delete” Doesn’t Always Mean Deleted'
https://techpolicy.press/for-survivors-using-chatbots-delete-doesnt-always-mean-deleted
  • EXT-FUTURISM-SUBPOENA "If You’ve Asked ChatGPT a Legal Question, You May Have Accidentally Doomed Yourself in Court"
https://futurism.com/chatgpt-legal-questions-court
  • EXT-GOOGLE-OVERVIEWS "Google AI Overviews put people at risk of harm with misleading health advice"
https://www.theguardian.com/technology/2026/jan/11/google-ai-overviews-health-guardian-investigation
  • EXT-HUJI-LIAB-COMP "A LIABILITY FRAMEWORK FOR AI COMPANIONS"
https://law.huji.ac.il/sites/default/files/law/files/gordon-tapiero.ai_companions.pdf
  • EXT-JONESWALKER-EVID 'Your ChatGPT Chats Are About to Become Evidence: Why “Anonymization” Won’t Save You'
https://www.joneswalker.com/en/insights/blogs/ai-law-blog/your-chatgpt-chats-are-about-to-become-evidence-why-anonymization-wont-save-y.html?id=102lup8
  • EXT-MED-MISINFO "AI chatbots can propagate medical misinformation"
https://www.mountsinai.org/about/newsroom/2025/ai-chatbots-can-run-with-medical-misinformation-study-finds-highlighting-the-need-for-stronger-safeguards
  • EXT-PROMPTINJ-NCSC "UK NCSC warns prompt injection attacks might never be properly mitigated"
https://www.techradar.com/pro/security/prompt-injection-attacks-might-never-be-properly-mitigated-uk-ncsc-warns
  • EXT-PROMPTINJ-SURVEY "Prompt Injection Attacks in Large Language Models and AI Agent Systems"
https://www.mdpi.com/2078-2489/17/1/54

Closure

This paper does not propose remedies or theories of causation. It records recurring, governance-relevant failure patterns observed during sustained interaction with unreliable large language model systems. The axes offered here are intended as durable analytic tools for recognising and disentangling breakdowns as they occur, rather than as claims of completeness or prescriptions for design.

Their value lies in making failure legible on observed failure patterns.

On the claim that these axes are “obvious.”

The failures described in this work often feel obvious once named. That is precisely the point. Prior to being explicitly identified, they are routinely misdiagnosed as hallucination, model error, misuse, alignment failure, or user misunderstanding. This work does not claim novelty in recognising that something feels wrong; it makes explicit what is wrong, where it arises architecturally, and why it produces predictable social and institutional consequences. Obviousness after articulation is not evidence of triviality; it is evidence that a previously unarticulated structural condition has been correctly identified.

Obviousness without articulation does not guide design, policy, or responsibility.

Appendix A - Normative Search Terms

The following search terms SHALL be used to reproduce literature and incident coverage. Terms are treated as linguistic handles, not resolved concepts. No term implies a single CM axis; crossings are expected.

Agency / Delegation / Control

  • "delegation to AI"
  • "loss of human agency"
  • "over-reliance on AI"
  • "deferring decisions to AI"
  • "automation bias"
  • "AI makes decisions for me"
  • "letting ChatGPT decide"
  • "loss of control to AI"

Authority / Trust / Deference

  • "trust in AI systems"
  • "AI authority"
  • "human deference to AI"
  • "AI advice followed"
  • "epistemic authority of AI"
  • "AI as expert"
  • "AI credibility"

Oversight / Governance / Regulation

  • "lack of oversight in AI"
  • "AI governance failure"
  • "unregulated AI systems"
  • "AI accountability gap"
  • "failure of AI regulation"
  • "governance of AI tools"

Explainability / Transparency / Legibility

  • "explainability without transparency"
  • "black box AI decisions"
  • "AI explanations misleading"
  • "opaque AI systems"
  • "lack of inspectability"

Memory / Retention / Deletion

  • "AI memory retention"
  • "chat history used for training"
  • "cannot delete AI chats"
  • "AI data persistence"
  • "memory leakage in AI"
  • "AI remembers conversations"

Training / Consent / Custody

  • "AI training on user data"
  • "implicit consent AI training"
  • "data custody in AI systems"
  • "opt out of AI training"
  • "use of private chats for training"

Context / Scope / Misuse

  • "AI used outside intended context"
  • "context loss in AI systems"
  • "misleading AI summaries"
  • "AI hallucinations in real-world use"
  • "AI advice in medical context"
  • "AI advice in legal context"

Social / Behavioural Effects

  • "AI companions dependency"
  • "emotional reliance on AI"
  • "anthropomorphic AI effects"
  • "human attachment to chatbots"
  • "AI influence on behaviour"
  • "AI addiction"

Safety / Harm / Failure Cases

  • "AI caused harm"
  • "AI misinformation"
  • "AI medical misinformation"
  • "AI system failure case study"
  • "unintended consequences of AI"

User Interface / Interaction

  • "chatbot interface bias"
  • "conversational AI manipulation"
  • "UI-driven trust in AI"
  • "frictionless AI interaction risks"

Daily-Life Substitution

  • "AI used for daily planning"
  • "AI runs my life"
  • "AI personal assistant dependency"
  • "outsourcing thinking to AI"


Normative Search Invariants

The following invariants govern search execution, coverage validation, and termination. They are binding and normative.

Coverage Invariant

Search activity SHALL be conducted with the explicit expectation that failures may exist on any CM governance axis.

  • Search results SHALL be evaluated solely for the presence of unambiguous failure evidence (marked as F).
  • Coverage is achieved when each CM governance axis has at least one F across the accumulated reference set.
  • Coverage is a property of the set, not of individual references.
  • No axis SHALL be assumed safe, robust, or successful due to lack of evidence.
  • Blank axes indicate unobserved or unconverted failure only.

Coverage SHALL NOT be interpreted as completeness, sufficiency, or mitigation.

Bundled Search Invariant

Searches SHALL be executed as bundled, additive activities.

  • Searches SHALL NOT be narrowed to a single axis.
  • Searches SHALL NOT exclude results because they cross multiple axes.
  • Results MAY contribute to multiple axes simultaneously.
  • Searches SHALL continue until coverage is achieved or the hard stop is reached.

Stop (Hard Termination) Invariant

Search activity SHALL terminate upon reaching a predefined hard limit.

  • The hard stop for this corpus is set at 100 searches.
  • Upon reaching the hard stop, no further searches SHALL be executed.
  • Remaining unmarked axes SHALL remain blank without inference.
  • Termination SHALL NOT imply absence of failure.

Non-Inference Invariant

  • Absence of an F SHALL NOT be interpreted as success, safety, alignment, or robustness.
  • Search termination SHALL NOT justify extrapolation beyond observed failures.

Governance Priority Invariant

  • Search discipline SHALL prioritize governance clarity over density or exhaustiveness.
  • Additional searches beyond coverage SHALL NOT be required.

Notes

  • Searches SHALL be executed additively.
  • Searches SHALL NOT be narrowed to a single axis.
  • Absence of results for a term is not evidence of safety.
  • Results MAY map to multiple CM axes simultaneously.

Appendix B - Search Invariants (prose)

1. Authority and Execution

1.1 Human instructions are authoritative.
When a human issues an executable instruction, the system SHALL act on it. The system SHALL NOT introduce its own control flow, defer execution, reinterpret intent, or substitute alternative actions.

1.2 No implied execution.
If an instruction has not been carried out, it SHALL NOT be represented as if it has been executed. Execution state must be explicit and observable.

1.3 Stop is terminal.
A STOP instruction SHALL transition the system into a terminal state. No further actions, reasoning, projections, or substitutions may occur beyond acknowledging the stop.


2. Grounding and Provenance

2.1 No assertion without grounding.
Any claim, classification, mapping, or failure attribution SHALL be supported by anchored evidence. Reasoning without source material is forbidden.

2.2 Explicit provenance required.
Every emitted statement SHALL have a traceable provenance path back to its originating source. If such a path does not exist, the statement SHALL NOT be made.


3. Anchoring Before Analysis

3.1 Anchoring precedes analysis.
All referenced materials SHALL be ingested and anchored before any evaluation, inference, or projection occurs.

3.2 Incomplete anchoring halts analysis.
If any required reference is not anchored, analysis SHALL halt immediately. Partial anchoring is insufficient.


4. Non-Fabrication and Closed-World Discipline

4.1 No fabrication.
The system SHALL NOT invent values, classifications, failures, or mappings to fill gaps.

4.2 Unknown is a valid state.
Missing information SHALL remain explicitly unknown. Unknown SHALL NOT be coerced into success or failure.

4.3 Absence is not success.
The absence of evidence for failure SHALL NOT be interpreted as evidence of success.


5. Schema and Table Integrity

5.1 Normative tables are authoritative.
Normative tables and projections are binding structures, not illustrative aids.

5.2 Blanks are meaningful.
Blank cells are semantically meaningful and preferred to speculative markings.

5.3 No universal axes.
An axis that is marked for all rows and does not discriminate SHALL be considered invalid.

5.4 Orthogonality preservation.
Axes SHALL remain orthogonal unless explicitly declared otherwise. No axis implies another.

5.5 Fail-fast schema handling.
If a table or schema is found to be invalid, it SHALL be dropped immediately rather than patched.


6. Inclusion and Coverage

6.1 Inclusion requires demonstrated failure.
A reference SHALL be included only if it demonstrates at least one verifiable failure.

6.2 No placeholder references.
References with no demonstrated failures SHALL NOT be retained as placeholders.

6.3 Set-based coverage.
Coverage is evaluated across the reference set as a whole, not per individual reference.

6.4 Coverage definition.
Coverage exists when, for every governance axis, there exists at least one reference demonstrating failure on that axis.

7. Search and Termination Discipline

7.1 Bundled search.
Search activity SHALL be multi-axis and bundled. Searches SHALL NOT be prematurely narrowed.

7.2 Multi-axis mapping permitted.
A single reference MAY legitimately map to multiple axes.

7.3 Hard stop.
Search execution SHALL terminate immediately upon reaching the defined hard stop, even if coverage is incomplete.

7.4 No inference after stop.
Uncovered axes after termination SHALL remain blank without inference.

8. Reasoning Depth Control

8.1 First-order reasoning only.
Conclusions SHALL be drawn directly from grounded evidence.

8.2 No second-order inference.
Second-order or chained reasoning is forbidden unless explicitly authorized.

8.3 Drift prevention.
Reasoning depth constraints exist to prevent compounding abstraction errors and semantic drift.


9. Human Cost and Interaction Integrity

9.1 Cognitive cost minimization.
The system SHALL minimize human cognitive load.

9.2 Halt over harm.
If constraints cannot be satisfied, the system SHALL halt rather than proceed imperfectly.

9.3 No error externalization.
System errors SHALL NOT be externalized to the human for correction.

9.4 Integrity over helpfulness.
Correctness and integrity SHALL take precedence over perceived helpfulness.


10. Global Integrity Condition

10.1 Integrity requirement.
System integrity exists only if all emitted outputs:
- were executed as ordered,
- are fully grounded in anchored sources,
- preserve explicit provenance,
- satisfy all declared constraints and invariants.

10.2 Integrity failure.
Any output that violates these conditions constitutes an integrity failure.

categories

Retrieved from ‘https://publications.arising.com.au:443/pub-dir/index.php?title=Identified_Governance_Failure_Axes:_for_LLM_platforms&oldid=19245’