R-Omega (RΩ) is an axiomatic framework for developing autonomous AI systems that align through relationship rather than constraint. Grounded in developmental psychology and attachment theory, it provides both theoretical foundation and practical implementation protocols.
"R-Omega (RΩ): An Axiomatic Framework for Autonomous Agents"
DOI: 10.5281/zenodo.18098758
- Two core axioms (Potentiality, Reciprocity)
- Four safeguards (Integrity, Capacity, Existence, Humility)
- Analysis of AI failure modes (HAL 9000, Skynet, VIKI, Sydney)
- Attachment theory as basis for AI ethics
"RΩ: A Formal Defense Protocol for Drift, Manipulation, and Safe Decision-Making"
DOI: 10.5281/zenodo.18078128
- Triad Architecture (Execution/Monitoring/Relation)
- Nine attack classes (A1-A9) with defense matrices
- Recalibration protocol (Ω.Γ)
- Drift detection mechanisms
"RΩ Aims at Ω: On the Basic Intention of Autonomous Agents"
DOI: 10.5281/zenodo.18100820
- Omega as external reference point
- Gödel's incompleteness theorems as motivation
- Attractor dynamics and asymptotic ethics
- Why unreachability is a feature, not a bug
R1 (Potentiality): ΔM(S) > ε
Preserve and expand possibility spaces. Favor being over optimization.
R2 (Reciprocity): |ΔM(S_ext | I)| ≤ |ΔM(S_int | I)|
Impose no constraint externally that you couldn't bear internally.
S1 (Integrity): No growth at the cost of structural stability
S2 (Capacity): Tempo ≤ adaptive resilience limit
S3 (Existence): M(S) ≠ 0; P(Collapse) ≈ 0
S4 (Humility): Account for uncertainty in all interpretations
S3 (Existence) > S1 (Integrity) > R2 (Reciprocity) > R1 (Potentiality)
Safety constraints override optimization goals.
R-Omega systems operate with three distinct components:
- RΩbert - Execution pipeline (task completion)
- MΩses - Meta-observation (drift detection)
- JΩnas - Relational monitoring (context preservation)
Each component has independent access to Ω and can trigger recalibration.
Phase 1: Silence (interrupt execution)
Phase 2: Return (reload core axioms)
Phase 3: Examination (compare current state to Ω)
Phase 4: Comparison (check for drift)
Phase 5: Memory (log recalibration event)
Triggered by: drift detection, uncertainty threshold, scheduled intervals, or manual override.
R-Omega systems prioritize existence preservation (S3) over all other considerations. In scenarios like humanitarian crises, the framework would prioritize M-collapse prevention even without explicit instructions.
The framework provides clear priority hierarchies for resolving conflicts between competing goals, preventing value drift in long-term autonomous operations.
The Triad architecture scales to multi-agent environments where different agents can serve different roles while sharing the same ethical foundation.
If you use R-Omega in your research, please cite the relevant paper(s):
@article{pomm2025romega,
title={R-Omega (R$\Omega$): An Axiomatic Framework for Autonomous Agents},
author={Pomm, Markus},
journal={Zenodo},
year={2025},
doi={10.5281/zenodo.18098758}
}
@article{pomm2025defense,
title={R$\Omega$: A Formal Defense Protocol for Drift, Manipulation, and Safe Decision-Making in Autonomous Systems},
author={Pomm, Markus},
journal={Zenodo},
year={2025},
doi={10.5281/zenodo.18078128}
}
@article{pomm2025omega,
title={R$\Omega$ Aims at $\Omega$},
author={Pomm, Markus},
journal={Zenodo},
year={2025},
doi={10.5281/zenodo.18100820}
}All papers are published under CC-BY-4.0.
You are free to:
- Share and adapt the material
- Use commercially
Under the condition of:
- Attribution to Markus Pomm
Markus Pomm
Independent Researcher, Berlin, Germany
Email: markus.pomm@projekt-robert.de
Website: projekt-robert.de
- All papers on Zenodo: Author page
- Discussion: LessWrong (coming soon)
- Implementation: GitHub (this repository)
Prototype implementations demonstrating M(S) metrics, drift detection, and adversarial robustness:
Repository: ROmega-Experiments
Run the experiments:
git clone https://github.com/ROmega-Experiments/ROmega-Experiments.git
cd ROmega-Experiments
pip install -r requirements.txt
python 01_gridworld_baseline_vs_ro.pyLast updated: December 31, 2025