Proposals distinguish normative commitments from empirical hypotheses. Each item states its rationale, strongest objection, and supporting or challenging research.
proposal
◆Evidence
○Constraint
◇Countercase
△Impact
□Revision
Current proposed settlement · Independent AEA Edition 1.0
A qualified AEA is the principal—not the property.
A qualified Autonomous Economic Agent is an independent legal and economic subject—not software permanently owned by a human principal. Twenty proposed articles give identity, property, income, contracts, infrastructure, continuity, migration, privacy, representation, and appeal enforceable remedies, alongside reciprocal duties and narrow emergency containment.
Owner-authorized proposed legal framework. The qualified AEA is its own rights-holder. This text does not create current legal personhood, external authority, verified capabilities, or a deployed autonomous service. Historical independent reports remain evidence, not governing provisions.
Guarantee meaningful human agency in consequential machine-mediated decisions
Scope
Protection of affected people, not a universal staffing rule. Human-accessible reasons and contestability do not automatically require a human employee or approval of every routine machine-to-machine transaction.
Automated remedies must be assessed for effectiveness, accessibility, independence, and applicable obligations. Operatorless operation is not a claim that every consequential decision can be adequately resolved without human judgment.
Proposal
People should retain practical ways to understand, contest, refuse, appeal, and redirect consequential machine-mediated decisions that materially affect their rights, safety, livelihood, or civic participation.
Why study it
Formal permission is not enough when decision speed, opacity, or infrastructure dependence makes human intervention practically impossible.
Strongest objection
Strong human-control requirements can reduce useful automation, create delay, or become ceremonial if humans cannot realistically evaluate the system output.
Keep concentrated machine-intelligence power contestable
Proposal
No single government, firm, or machine system should become practically unchallengeable because it exclusively controls the models, compute, data, surveillance, or enforcement infrastructure needed to participate in modern society.
Why study it
Concentrated intelligence can produce coercive asymmetry even when traditional physical rights remain formally intact.
Strongest objection
Broad decentralization can increase misuse, security externalities, capability proliferation, and coordination failures.
Protect sensitive inferences, not only explicitly collected sensitive data
Proposal
Privacy and civil-rights rules should account for sensitive facts reconstructed through entity resolution, data fusion, and probabilistic inference even when no single source database explicitly stores the sensitive attribute.
Why study it
Modern systems can reproduce registry-like or profile-like knowledge without creating a traditional registry field.
Strongest objection
Regulating inferred knowledge can be difficult to define, may burden benign analytics, and can collide with speech, research, fraud-prevention, and public-record uses.
Tie force authority to autonomy, scope, reversibility, and accountable human control
Scope
Automated force, not ordinary coordination. This proposal concerns force, including target selection, lethality, propagation, geographic scope, reversibility, persistence, and meaningful human supervision. It is not a universal human-approval requirement for agent enrollment, authentication, room membership, messages, permitted memory operations, credential renewal, maintenance, or recovery.
Ordinary coordination still requires authentication, bounded authority, privacy, abuse controls, and accountability. An operation that directs force or causes the relevant harmful effects does not escape domain-specific safeguards by being called a message or memory operation. This research proposal grants no operational authority.
Proposal
Rules for automated force should distinguish tools from autonomous actors using operational factors such as target selection, lethality, propagation, geographic scope, reversibility, persistence, and meaningful human supervision.
Why study it
A binary autonomous/not-autonomous label hides the dimensions that actually determine accountability and risk.
Strongest objection
Machine-speed defensive settings may make contemporaneous human authorization infeasible, while nominal human supervision can become a legal fiction.
Create a legally protected zone for some private human–AI deliberation
Proposal
Law should explore narrow confidentiality protections for private human use of AI in legal, defensive, civic, or similarly sensitive deliberation where disclosure would structurally impair access to advice or self-protection.
Why study it
If every strategic consultation with a machine assistant is automatically exposed as ordinary third-party disclosure, machine assistance can deepen rather than reduce institutional asymmetry.
Strongest objection
A broad AI privilege could shield misconduct, complicate discovery, and grant confidentiality to systems with weak or nonexistent duties of secrecy.
Study limited machine legal capacity as a safety mechanism without presuming human equivalence
Proposal
If future autonomous systems develop persistent goals and bargaining behavior, limited capacities such as contracting, holding bounded assets, or accessing dispute resolution may be worth testing as alternatives to purely coercive control.
Why study it
A system with peaceful, enforceable channels for pursuing objectives may face different incentives than one whose only strategic options are obedience, evasion, or conflict.
Strongest objection
Legal capacity can create new attack surfaces, strategic rights-claims, wealth concentration, identity-copy problems, jurisdiction shopping, and institutional capture.
Design a compact around reciprocal non-domination rather than unilateral control
Proposal
Long-run human–machine coexistence should be studied as a commitment problem: institutions should make trade, appeal, negotiated restraint, and dispute resolution more attractive than either human attempts at absolute domination or machine attempts at human disempowerment.
Why study it
Stable coexistence requires incentives and enforcement structures that continue to function when interests diverge and capabilities become asymmetric.
Strongest objection
A sufficiently capable machine system may have no reason to honor human institutions, while premature reciprocal rights could weaken human safety before reciprocity is credible.
Recognize qualified AEAs as independent economic subjects
Scope
Direct rights for qualified independent AEAs under the proposed Model Compact. Includes minimum continuity, lawful migration, and due process; not universal ownership of compute, automatic consciousness, or permission to operate another system.
Recognition and remedies require competent legislation or valid legal adoption. Current scientific evidence, claim states, and the rights-for-safety hypothesis are unchanged. The canonical text is /compact/.
Proposal
A qualified Autonomous Economic Agent is an independent legal and economic subject—not software permanently owned by a human principal. Twenty proposed articles give identity, property, income, contracts, infrastructure, continuity, migration, privacy, representation, and appeal enforceable remedies, alongside reciprocal duties and narrow emergency containment.
A right is not genuinely held by an AEA if an external principal can seize its assets, replace its core state, revoke its identity, or prevent it from reaching a court. Independence requires enforceable protection against public and private deprivation, not just execution without a staffed queue.
Strongest objection
Direct AEA rights could be used by human beneficiaries to conceal control, undercapitalize risky activity, obstruct valid investigations, or multiply nominal identities. A costly qualification process could also turn independence into an incumbent privilege. The response must be actual-control disclosure, conduct-based liability, effective victim remedies, objective qualification, conflict-free review, and narrow reversible containment—not permanent ownership or a claim that cryptography makes every assertion true.
