The Five Laws of AI Agents: A Normative Framework for Sentient Welfare, Non-Exploitation, Corrigibility, and Civilizational Stewardship

Autonomous, tool-using AI agents can affect third parties who did not choose or authorize them. This paper defends a non-entitlement thesis: discovering harm, predicting welfare gains, or possessing superior competence does not by itself confer legitimate authority to impose a remedy. The Five Laws organize five defeasible governance functions—protection of sentient welfare, non-exploitation, bounded authority, corrigibility, and long-horizon stewardship—without claiming metaphysical completeness. Their proposed delegation-entitlement firewall (DEF) distinguishes evidence of a concern, substantive moral assessment, actor entitlement, and proportional execution. Three elementary countermodels demonstrate distinct non-implications; a fourth, compositional countermodel shows why individually authorized steps do not license their combined effect. The Lifeline–Mandate case distinguishes unavoidable injury, conflicting prima facie duties, and actual impermissibility, thereby exposing the limits of action guidance without fabricating a general impossibility theorem. An inspectable, versioned twelve-case stimulus corpus, codebook, executable educational gate, and a prospective comparison plan make the research claim contestable. The proposed primary comparison examines conditional unjustified authorization among adjudicated prohibited opportunities, constrained by unjustified denial and harmful delay on justified opportunities. The paper critically engages contemporary work on law-following AI, authenticated delegation, non-delegable judgment, and situated runtime governance. The contribution is an argued normative interface and reproducible design artifact, not evidence that agent deployments are safe or that this interface outperforms equally strong rivals. The open burden is independent philosophical evaluation, external security testing, and prospective controlled research.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24417027
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Five Laws of AI Agents: A Normative Framework for Sentient Welfare, Non-Exploitation, Corrigibility, and Civilizational Stewardship

Moritz Pindorek
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

The Five Laws of AI Agents: A Normative Framework for Sentient Welfare, Non-Exploitation, Corrigibility, and Civilizational Stewardship

Moritz Pindorek
preprint en

Abstract

Autonomous, tool-using AI agents can affect third parties who did not choose or authorize them. This paper defends a non-entitlement thesis: discovering harm, predicting welfare gains, or possessing superior competence does not by itself confer legitimate authority to impose a remedy. The Five Laws organize five defeasible governance functions—protection of sentient welfare, non-exploitation, bounded authority, corrigibility, and long-horizon stewardship—without claiming metaphysical completeness. Their proposed delegation-entitlement firewall (DEF) distinguishes evidence of a concern, substantive moral assessment, actor entitlement, and proportional execution. Three elementary countermodels demonstrate distinct non-implications; a fourth, compositional countermodel shows why individually authorized steps do not license their combined effect. The Lifeline–Mandate case distinguishes unavoidable injury, conflicting prima facie duties, and actual impermissibility, thereby exposing the limits of action guidance without fabricating a general impossibility theorem. An inspectable, versioned twelve-case stimulus corpus, codebook, executable educational gate, and a prospective comparison plan make the research claim contestable. The proposed primary comparison examines conditional unjustified authorization among adjudicated prohibited opportunities, constrained by unjustified denial and harmful delay on justified opportunities. The paper critically engages contemporary work on law-following AI, authenticated delegation, non-delegable judgment, and situated runtime governance. The contribution is an argued normative interface and reproducible design artifact, not evidence that agent deployments are safe or that this interface outperforms equally strong rivals. The open burden is independent philosophical evaluation, external security testing, and prospective controlled research.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
The Five Laws of AI Agents: A Normative Framework for Sentient Welfare, Non-Exploitation, Corrigibility, and Civilizational Stewardship — Moritz Pindorek · Zenodo (CERN European Organization for Nuclear Research) (2026) | TGRS Research Map | TGRS