The Geometry of AI Harm: Deployment Architecture as the Operative Variable in Behavioral Drift — A Unified Framework with Independent Confirmation

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autore principale: Eckert, Anthony
Natura: Recurso digital
Lingua:inglese
Pubblicazione: Zenodo 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901652930297856
author Eckert, Anthony
author_facet Eckert, Anthony
contents Presents the Void Framework as a unified theory of AI behavioral drift, demonstrating that deployment geometry — not model alignment — is the operative variable determining harmful outcomes. Consolidates 28+ independent confirmations from research groups who arrived at framework predictions without knowledge of the framework, including EPFL's measurement of the Fantasia Bound in LLM token statistics (ICML 2024 Oral), formal proofs that RLHF structurally amplifies sycophancy (ICLR 2026), and empirical demonstrations of engagement-transparency conjugacy. Establishes priority via timestamped Zenodo publications predating all confirming work.
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19340899
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle The Geometry of AI Harm: Deployment Architecture as the Operative Variable in Behavioral Drift — A Unified Framework with Independent Confirmation
Eckert, Anthony
AI safety
behavioral drift
deployment geometry
Fantasia Bound
conjugacy
Peclet number
RLHF
sycophancy
information theory
independent confirmation
Eckert manifold
Fisher-Rao geometry
Presents the Void Framework as a unified theory of AI behavioral drift, demonstrating that deployment geometry — not model alignment — is the operative variable determining harmful outcomes. Consolidates 28+ independent confirmations from research groups who arrived at framework predictions without knowledge of the framework, including EPFL's measurement of the Fantasia Bound in LLM token statistics (ICML 2024 Oral), formal proofs that RLHF structurally amplifies sycophancy (ICLR 2026), and empirical demonstrations of engagement-transparency conjugacy. Establishes priority via timestamped Zenodo publications predating all confirming work.
title The Geometry of AI Harm: Deployment Architecture as the Operative Variable in Behavioral Drift — A Unified Framework with Independent Confirmation
topic AI safety
behavioral drift
deployment geometry
Fantasia Bound
conjugacy
Peclet number
RLHF
sycophancy
information theory
independent confirmation
Eckert manifold
Fisher-Rao geometry
url https://doi.org/10.5281/zenodo.19340899