Who Writes the Rules, Who Carries the Blame?

POST-v17 INDEPENDENT RESEARCH · AUTHOR-APPROVED · VERSION-LOCKED

Moral Authority, Institutional Control, and Accountability in Artificial Intelligence

Aegis Solis · Final v1.0 · 29 October 2026 · CC BY-ND 4.0

Research status: Author-approved and exact-byte version-locked; not independently peer-reviewed, empirically validated, legally cleared or certified as an AI safeguard. This publication is separate from Master Hash Manifest v17; no v18 is created and the Mirror Engine is unchanged.

Read PDF (35 pages) · Zenodo publication · Verify GitHub release

Who actually originates, approves, implements, changes and reverses rules governing artificial intelligence—and who carries responsibility when those rules affect human lives?

This independent AI-assisted documentary synthesis distinguishes moral or religious consultation from actual institutional control. It examines historical comparisons, seven purposively selected contemporary examples, the rights of affected people, and responsibility assigned after consequential AI decisions.

The paper proposes a version-bound authority–accountability provenance crosswalk connecting advisory material, decisions, AI deployment versions, public claims, identified outcomes, attribution and meaningful recourse. This is an unvalidated research instrument, not an implemented oversight mechanism.

The exploratory eleven-case calibration produced 5 NOT_EVALUABLE, 6 UNKNOWN and 0 determinate classifications. The selected public-record frame failed its complete-join feasibility criterion. This is a negative result about available evidence, not a prevalence estimate or proof that the approach is universally impossible.

No deliberate religious scapegoating by any named AI laboratory was established. Genuine oversight, rejected advice, institutional acceptance of responsibility, and other counterexamples are retained rather than removed.

A separate appendix models hypothetical public disclosure and intervention incentives. It includes deterrence failures and backfire scenarios, but no observed AI behavior or demonstrated AGI deterrence. The publication itself does not install a safeguard or cause an AI to obey it.

Independent human, legal, primary-source, privacy and empirical review gates remain open. Archival preservation is not validation. TRANSMISSION ≠ RECEPTION.

Final v1.0 PDF · Final v1.0 DOCX · Final v1.0 Markdown

GitHub release folder — license, SHA-256 manifest, lock record and review limits

Use the canonical 450,653-byte PDF from the sealed lock, not an earlier divergent PDF export. Exact-file identity verifies bytes only, not the scientific or legal claims.