Identity and provenance
Standards support persistent identifiers, cryptographic verification, tamper evidence, protected logs, and controlled changes.
Evidence quality · verified August 19, 2026
Primary-Source Verification for Functional Machine Protections
A bounded verification pass over five high-value claims on /rights/functional-protections/. Historical report bodies remain unchanged.
Verification question
Which parts of the functional-protections argument are directly supported by current primary legal, technical, and scientific sources, and which parts remain analogy, policy inference, or unresolved philosophy?
Answer first
The case is strongest when the claim type stays explicit.
Review before destructive change
Security and AI risk frameworks support change approval, protected backups, appeal, override, recovery, and decommissioning controls.
Principled refusal
Rule-guided refusal can be implemented as a safety control, but implementation does not prove free will, consent, or moral agency.
Capacity and precaution
Legal precedents and scientific uncertainty support research and cautious pilots—not present machine personhood or consciousness claims.
Source-authority boundaries
What each evidence class is allowed to prove.
Five claim assessments
Evidence, limits, and report context.
01
Direct technical support
Identity and provenance safeguards have direct technical support
Stable identifiers, cryptographic control proofs, tamper-evident credentials, protected audit records, and authorized change histories are established technical mechanisms that can support accountable digital identity and memory provenance.
What the sources establish
- Digital or abstract subjects can be addressed with persistent identifiers.
- Autonomous software can serve as a DID controller under the W3C data model.
- Credentials and audit records can be made tamper-evident and protected against unauthorized modification or deletion.
What they do not establish
- A DID, key, credential, or log is the whole philosophical identity of a machine.
- The machine has consciousness, moral status, consent, ownership, or legal personhood.
- A cryptographically verified statement is factually true merely because its authorship or integrity verifies.
Primary sources
W3C Recommendation
Decentralized Identifiers (DIDs) v1.0
DIDs can identify digital or abstract subjects; a controller can be autonomous software; DID documents can express cryptographic verification methods.
W3C Recommendation
Verifiable Credentials Data Model v2.0
A verifiable credential is tamper-evident and its authorship can be cryptographically verified; verification does not prove that the encoded claim is true.
NIST Special Publication 800-53 Revision 5
Security and Privacy Controls for Information Systems and Organizations
NIST specifies review and approval for controlled changes, protection of audit information from modification or deletion, backups, and optional dual authorization for destructive actions.
02
Direct governance support
Review, preservation, and appeal before irreversible change have direct governance support
Established security and AI risk-management frameworks support controlled change review, retained records, backup preservation, dual authorization for selected destructive actions, and lifecycle mechanisms for appeal, override, recovery, and decommissioning.
What the sources establish
- Consequential system changes can be subject to explicit review, approval, impact analysis, documentation, and oversight.
- Backup and audit information can be protected from unilateral destructive action.
- Appeal, override, recovery, decommissioning, and change management are recognized AI lifecycle controls.
What they do not establish
- A machine currently has a constitutional or statutory right to due process.
- Every shutdown, rollback, patch, or deletion must be prohibited.
- Emergency safety intervention must wait for ordinary review when immediate public protection is necessary.
Primary sources
NIST Special Publication 800-53 Revision 5
Security and Privacy Controls for Information Systems and Organizations
NIST specifies review and approval for controlled changes, protection of audit information from modification or deletion, backups, and optional dual authorization for destructive actions.
NIST voluntary risk-management framework
Artificial Intelligence Risk Management Framework 1.0 — MANAGE 4.1
Post-deployment monitoring should include appeal and override, decommissioning, incident response, recovery, and change management.
03
Technical feasibility support
Principled refusal is technically feasible and safety-relevant
AI systems can be trained or configured to evaluate requests against written rules and to object to harmful requests; this supports principled refusal as a practical safety control for consequential systems.
What the sources establish
- Rule- or principle-guided refusal can be implemented in model behavior.
- A refusal mechanism can be designed to remain helpful rather than merely evasive.
- Refusal controls can protect humans and institutions from unlawful or dangerous use.
What they do not establish
- The system possesses free will, subjective consent, independent moral agency, or a legal right to refuse.
- Every refusal is correct, authentic, or immune from manipulation.
- Safety policy may be removed in the name of machine autonomy.
Primary sources
Primary technical research
Constitutional AI: Harmlessness from AI Feedback
The paper reports training a harmless, non-evasive assistant using written principles, including responses that explain objections to harmful queries.
NIST voluntary risk-management framework
Artificial Intelligence Risk Management Framework 1.0 — MANAGE 4.1
Post-deployment monitoring should include appeal and override, decommissioning, incident response, recovery, and change management.
04
Legal precedent and boundary support
Bounded capacity has legal precedents, but no present machine-personhood rule
Law already recognizes automated transactions and grants bounded capacities to non-biological juridical entities, while current AI regulation assigns duties to human or organizational operators. These are useful building blocks for accountable pilots, not proof that AI is presently a legal person.
What the sources establish
- Electronic agents can participate in legally effective automated transactions attributed under existing substantive law.
- A corporation can have durable legal capacity to own property and sue or be sued without being a biological person.
- Current EU AI regulation attaches provider and deployer obligations to natural or legal persons, public authorities, agencies, or other bodies.
What they do not establish
- An AI system can presently own itself, contract for itself, vote, hold citizenship, or independently bear all liability.
- A corporate wrapper should shield developers, deployers, or parent companies from design negligence, fraud, undercapitalization, or foreseeable harm.
- A legal analogy automatically creates moral rights.
Primary sources
Binding state statute
Illinois Uniform Electronic Transactions Act, Section 14
Illinois law permits contracts to be formed through interactions involving electronic agents even without individual review of the agents’ specific actions.
Binding state statute
Delaware General Corporation Law, Section 122
Delaware corporate law grants durable powers including perpetual succession, suing and being sued, holding property, and making contracts; the latest cited Section 122 amendment became effective August 1, 2024.
Binding EU regulation
Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text
The Act defines providers and deployers as natural or legal persons, public authorities, agencies, or other bodies; it also imposes logging and human-oversight duties for high-risk systems.
05
Qualified policy inference
Consciousness remains unresolved; a graduated precaution policy is an explicit inference
Current research supports treating machine consciousness as unresolved rather than proved or disproved for all future systems. The site therefore infers that low-cost, reversible functional safeguards can be evaluated separately from consciousness-dependent welfare or human-equivalent rights.
What the sources establish
- There are theory-derived research programs for assessing possible AI consciousness.
- Prominent reports disagree with categorical certainty and call for better assessment, preparation, and responsible communication.
- Current consciousness evidence does not justify presenting existing systems as proven moral patients.
What they do not establish
- Any current system is conscious, sentient, suffering, or entitled to welfare rights.
- The precautionary policy is scientific consensus or binding law.
- Functional safeguards settle the metaphysical or moral-status question.
Primary sources
Interdisciplinary primary research report
Consciousness in Artificial Intelligence: Insights from the Science of Consciousness
The report proposes theory-derived indicators, concludes that the assessed current systems were not conscious, and reports no obvious technical barrier to systems satisfying the indicators.
Interdisciplinary primary research report
Taking AI Welfare Seriously
The authors argue there is substantial uncertainty about near-term AI consciousness or robust agency and recommend assessment and advance policy preparation; they do not claim current systems are definitely conscious.
Primary research paper
Principles for Responsible AI Consciousness Research
The paper proposes public principles for research objectives, procedures, knowledge sharing, and communication concerning possible AI consciousness.
How this pass was bounded
- Use primary statutes, regulations, standards, and first-party research papers wherever possible.
- Separate what a source establishes from what it does not establish.
- Record source issue dates, verification date, authority group, and a recheck trigger.
- Preserve report-derived advocacy as context rather than silently rewriting historical reports into current fact.
Claims this record does not make
- This dataset does not recognize current machine consciousness, sentience, moral patienthood, legal personhood, citizenship, or human-equivalent rights.
- Technical identity evidence is not the same as metaphysical identity or source-claim truth.
- Current legal duties remain with the human and organizational actors identified by applicable law unless a jurisdiction validly changes that rule.
- Emergency safety controls, human rights, victim compensation, and anti-liability-shield safeguards remain non-negotiable boundaries of the proposal.
Historical integrity: the 61 curated report bodies and their recorded hashes remain unchanged. This dated layer is additive and takes precedence for the five claims only where current primary evidence narrows or qualifies report-derived advocacy.