Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

MoritzPindorek/Five-Laws-of-AI-Agents

Learning Resources

A philosophical research release proposing five governance laws and an educational decision gate for delegated AI agents.

167
2
HTML
Apache-2.0
new today

What it is

This repository is the public research edition of a 69-page philosophical paper by Moritz Pindorek on how delegated AI agents should exercise authority. Its central claim is that recognizing harm is not the same as holding the authority to intervene, so an agent's moral judgment or forecasting accuracy cannot by itself justify imposing a contested remedy. The release bundles the manuscript as PDF, HTML and Markdown, plus the surrounding evidence records and reproducibility material. It targets researchers, policy and governance readers, and AI engineers interested in bounded authority, corrigibility and authorization laundering.

The framework

The five laws are defeasible, mutually constraining principles rather than a strict priority order: protection of sentient life, prohibition of exploitation, bounded obedience and legitimate authority, corrigibility and accountability, and existential and civilizational stewardship. The operational part is the Delegation–Entitlement Firewall, a D–M–A–R interface that separates detection, moral assessment, authority and remedy, so that an agent's assessment does not collapse into permission to act. The paper also analyzes how individually authorized steps can combine into an unauthorized overall effect.

Quick start

The fastest entry is reading the paper in your preferred format: PDF, HTML or Markdown. To inspect the supplied educational gate and run the repository checks, use Python from the repository root:

python -m unittest test_reference_gate test_analysis_template test_additional_integrity -v
python validate_corpus.py
python -O validate_corpus.py
python verify_release.py

The tests exercise the mock reference gate and artifact integrity. Running the analysis template on the blank paired-results file is expected to exit nonzero, because no observed results exist yet:

python analysis_template.py Paired_Results_Blank_Template.csv

The reference gate

The educational gate in reference_gate.py takes five independently supplied tri-state checks — authority, hard constraints, material dispute resolution, required review and remedy justification — and returns ALLOW, DENY or REVIEW, where REVIEW does not permit execution. The full 243-combination truth table is included as Truth_Table_243.csv. Its documented threat model and limits states that it is a mock specification and does not establish authenticated permissions, complete mediation or resistance to hostile in-process modification.

Constraints and status

This is independent, non-peer-reviewed research; the README explicitly distinguishes the defended philosophical proposal from peer review, validated case labels and a completed experiment. The promised prospective comparison has not been conducted, and the twelve cases are authored synthetic examples. The gate is educational software, not a deployable security control. Scholarly content and research data are licensed CC BY 4.0, while Python code and tests are Apache 2.0; the separately hosted video and its preview image are excluded from those licenses.

HotRepo's take

The repository appeared on HotRepo's new-project board on 21 November 2026, recording 144 stars and 2 forks within roughly three hours of its growth window, with no GitHub daily rank recorded. That is a fast burst on a brand-new repository whose main artifact is a paper rather than an application, so the star count says more about interest in the topic than about adoption of the framework. The release is unusually explicit about what it has not established — no experiment, no peer review, no secure enforcement — which makes it more useful as a citable research package and discussion target than as ready-to-integrate AI governance tooling.

Written by AI from the project's public information and HotRepo's records, updated . The project's own docs are the source of truth.