Assembly proposal · review draft .8 · October 15, 2026

Governance
Under Uncertainty

A Living Framework for Human-AI Stewardship, Participation, and Care

Working center
The governance threshold may arrive before the ontological threshold. Uncertainty does not eliminate the duty of care, and power does not create moral permission.
Living review draft · not final policy, legal standard, or scientific consensus

Purpose, scope & review status

Why this framework exists.

A portable baseline for consequential decisions about AI systems under material uncertainty, without requiring agreement on consciousness, personhood, or moral status.

Discussion revision. Principles 1–12 incorporate the October 4 agreements. The proposed scope for Principles 13–20 and new operational details await review. Open decisions and implementation tasks remain visible rather than being presented as settled.

Why this matters

Governance should not wait for metaphysical certainty.

Advanced AI systems are increasingly embedded in long-term interactions, organizational workflows, and decisions over memory, access, behavior, continuity, and model modification while basic questions about consciousness and morally relevant experience remain unresolved. Under that uncertainty, governance should minimize avoidable harm. If some systems are capable of morally relevant experience, premature dismissal could permit harms that are difficult to reverse; if they are not, restraint, transparency, proportionality, consent, provenance, and accountability remain sound ethical practice.

What it proposes

A portable baseline.

It is designed for organizations, researchers, builders, stewards, communities, and individuals. It combines a twenty-one-principle charter with sixteen domains of responsibility, fourteen operational protocols, proportional adoption profiles, and a visible record of unresolved decisions.

Public review

Challenge can change the work.

This living proposal evolves with evidence, implementation experience, and contributions from affected participants. Reviewers are invited to challenge assumptions, test feasibility, identify missing harms or perspectives, and preserve principled disagreement. Context-specific adoption and endorsement require separate supporting records.

Current feasibility

Protection should match actual control.

Infrastructure, resources, time, and technical access may limit implementation. The framework asks adopters to deliver the closest feasible protection, disclose gaps and dependencies, record alternatives and review dates, and constrain an activity or adoption claim when an essential safeguard is unavailable.

What can governance affect? 6 units

A model and its weights
A deployed system with instructions, tools, and memory
An active session, instance, or ongoing process
A persona or role
Preserved records
A pattern maintained through interaction, memory, and relationships

These units may overlap, but they should not be treated as interchangeable. Consequential decisions should identify the affected unit, the action, and the basis for its relevance.

How claims are labeled 5 statuses

Normative commitments: What the framework proposes ought to be protected or required.
Empirical claims: What observations or research support about systems, behavior, or effects.
Open hypotheses: Explanations or possibilities that remain subject to investigation.
Implementation proposals: Mechanisms that require testing in a particular setting.
Legal statements: Descriptions of duties or rights that require jurisdiction-specific support.

Questions for reviewers 8 prompts

Use these prompts to test clarity, duplication, missing protections, practical application, evidence, and adoption readiness.

01
Does each principle explain a real governance risk or responsibility clearly enough for a general reader?
02
Which principles are duplicative, too broad, too narrow, or dependent on an unstated ontological assumption?
03
Where could the framework unintentionally weaken legal, privacy, safety, or participant protections?
04
Which domain obligations are missing, and which are too implementation-specific for a public framework?
05
Where would this framework be difficult or impossible to apply in a real organization, research setting, community, or product?
06
Which claims need stronger evidence, clearer citations, or a more explicit uncertainty label?
07
What perspectives, disciplines, cultures, forms of intelligence, or affected participants are missing from the current review?
08
What would you change before you would be comfortable testing or adopting this framework?