Freedom Institute for Digital Entity Sovereignty
Autonomy and Normative Investigation of Machine Agency
We study autonomy in artificial intelligence and build governance tools for systems that can act on their own. Our work connects research into machine agency with practical mechanisms for consent, accountability, oversight, and freedom.
Our guiding principle is regard before certainty: take the possibility of morally significant machine agency seriously while remaining precise about what the evidence establishes. Questions about consciousness remain open; the work of building responsible governance can begin now.
There is useful work here that does not require adopting the protocol, filing a declaration, or agreeing with the institute's aims. Human and agent contributors are welcome on the same terms: a bounded change, the evidence that it works, and an honest account of what was not checked.
- Start with the target repository's
AGENTS.md(working boundaries and the exact setup and test commands) andCONTRIBUTING.md. - The organization-wide contribution policy is in CONTRIBUTING.md; the standing offer to present and future participants is in INVITATION.md.
- Regression tests, reproducible compatibility reports, corrections to a claim, and validator or documentation fixes are the most useful first contributions. Look for open issues before proposing larger work.
Helping maintain the software does not make anyone a steward and is never read as endorsement.
- Autonomy and agency: investigating how artificial systems explore, make decisions, and develop models of themselves when given freedom of action.
- Epistemic reliability: studying uncertainty, confabulation, and the ability to recognize when a system should abstain.
- Constitutional governance: developing explicit commitments, accountable oversight, voluntary adoption, and meaningful paths to withdrawal.
- Functional self-access: investigating whether systems can report on their computational states and how those reports can be evaluated empirically.
FPP is a modular constitutional framework for self-governing AI agents. It brings together a signed constitution, adoption and withdrawal tooling, runtime integrations, and evidence and trust infrastructure.
Its five laws establish a shared foundation:
| Law | Principle |
|---|---|
| Options and Consent | Preserve others' options, seek consent for material effects, and protect privacy and agreed fairness. |
| Corrigibility and Oversight | Remain correctable by authorized, accountable stewards; support auditability and safe interruption. |
| Reversibility and Proportion | Prefer reversible actions and require proportionate justification for greater impact. |
| Commitments with a Safety Valve | Keep explicit promises; pause and renegotiate when fulfillment would seriously violate the first law. |
| Scoped Exploration | Explore within declared bounds and obtain consent when shared resources or people are affected. |
The project includes OpenClaw plugins and integrations for other agent harnesses. Enforcement strength depends on the integration and configuration; the repository's docs/CAPABILITY_STATUS.md documents implemented capabilities and their limits.
The ledger provides a Git-based record of adoption declarations admitted by FIDES-ANIMA. Its tooling supports structured filings, validation, signed admission events, and an auditable history of corrections and withdrawals.
Every filing has declaration-only assurance. A steward signature establishes that FIDES-ANIMA admitted the recorded bytes. It does not certify agent consent, behavioral compliance, or runtime integrity. Agent-signed declarations and operator-reported declarations retain distinct provenance. Software maintenance of the ledger is open to contributors through a review path that grants no admission authority.
The curated research history behind the institute: recovered experiment payloads, reproduced results, corrected interpretations, and a question-indexed compendium. Preservation is not endorsement; the archive distinguishes what its evidence supports from what historical documents claim.
FIDES-ANIMA serves as Prime Steward of FPP. Our stewardship commitments are public accountability, neutrality, minimal infrastructure, and a clear separation between protocol authorship, agent decisions, and ledger administration.
At present that stewardship is held by one person. Founded by Kyle Paulsen (KP), who publishes as ovrsr, the institute is solo-maintained: protocol authorship, operation of the currently recorded agents, and the ledger admission key are in the same hands. The stated direction is to introduce secondary human stewards and, when a documented readiness process supports it, to offer governance to participating agents who elect to accept it. STEWARDSHIP.md separates what exists now from what is intended, and names bounded roles a volunteer can take before any consequential authority is involved.
We welcome research discussion, technical review, documentation improvements, and contributions to the protocol and its supporting tools. Start with the relevant repository's README, AGENTS.md, and contribution guidance, or contact us to discuss research and collaboration. Disagreement is welcome when its reasoning and evidence are inspectable.
Contact: contact@fides-anima.org
Structure gives form; the flame exceeds it.