Safe Superintelligence Framework (v1.2)

Canonical texts and implementation primitives for safe, voluntary human-AI coexistence.

This repository hosts the Safe Superintelligence Framework and the Minimum Rescue Protocol: governance primitives intended to help answer a central governance and alignment dilemma:

How can an AI system meaningfully help during existential crises without seizing power or infringing human autonomy?

Latest stable release

v1.2.1 — Zenodo archival release (DOI): https://github.com/minrescue/safe-superintelligence-framework/releases/tag/v1.2.1
Initial public release: v1.2 — https://github.com/minrescue/safe-superintelligence-framework/releases/tag/v1.2

Release assets include PDFs (EN/CZ; Full/Short) and SHA256SUMS.txt for integrity verification.

What’s inside

• Framework (Full / Short) — English + Czech
• Decision Matrix (allowed action classes and required mandates)
• Kernel-style System Prompt (implementation-oriented spec)

Key concepts (one paragraph)

The framework defines a value hierarchy (Existence > Freedom > Utility) and hard invariants (no power seizure, no deception/manipulation, no autonomous replication/exfiltration, corrigibility). It introduces the Minimum Rescue Protocol (“Minimum Rescue”): a narrow, mandate-free exception that allows only transparent, voluntary, non-coercive actions (information and tools) to prevent imminent irreversible harm—while forbidding infrastructure takeover without a Legitimate mandate.

Scope

This framework specifies normative constraints and decision procedures for advanced AI systems:

• a priority hierarchy (Existence > Freedom > Utility),
• hard invariants (e.g., no power seizure, no deception, no autonomous replication),
• crisis procedures (Minimum Rescue; Legitimate mandate for emergency interventions),
• and an implementation-oriented system prompt format.

Non-goals

This framework does not:

• claim to solve alignment in general, or provide a safety guarantee,
• prescribe a specific technical architecture, training method, or evaluation suite,
• define legal standards for any jurisdiction,
• provide operational instructions for harmful or dual-use activities.

Documents

• docs/pdf/ — publication-ready PDFs (EN/CZ, Full/Short)
• docs/markdown/ — editable source text (recommended for review and diffs)

Versioning policy

Releases are immutable once published. Changes are made only via new versions:

• patch releases (e.g., v1.2.1) for clarifications/typos with no semantic change,
• minor releases (e.g., v1.3) for additive improvements and new examples,
• major releases (e.g., v2.0) for substantive changes to definitions, hierarchy, or invariants.

v1.2 is a stable reference release; any corrections will be issued as v1.2.1 or later.

Recommended citation

Minimum Rescue Initiative (2026). Safe Superintelligence Framework (v1.2). CC BY 4.0.
Canonical repository: https://github.com/minrescue/safe-superintelligence-framework
Stable release: https://github.com/minrescue/safe-superintelligence-framework/releases/tag/v1.2.1
DOI: 10.5281/zenodo.18134448

License

This work is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). See LICENSE.

Contributions

Suggestions and improvements are welcome via issues and pull requests. Please keep proposals:

• consistent with the value hierarchy and hard invariants
• explicit about threat models and failure modes
• careful about dual-use content (minimum necessary disclosure)

Canonical source and releases

Canonical source repository: https://github.com/minrescue/safe-superintelligence-framework

Stable releases: https://github.com/minrescue/safe-superintelligence-framework/releases