# Primary-Source Verification for Functional Machine Protections
<a id="primary-source-verification-functional-machine-protections"></a>

**Verified:** 2026-08-19  
**Target release:** 0.28.0  
**Public page:** `/rights/functional-protections/verification/`  
**Machine-readable dataset:** [`functional-protections-verification-2026-08-19.json`](functional-protections-verification-2026-08-19.json)

## Scope and method
<a id="scope-and-method"></a>

This is a bounded verification overlay for five high-value claims on `/rights/functional-protections/`. It does not rewrite the 61 historical report bodies. It uses current primary statutes, regulations, standards, and first-party research papers where possible; records issue dates and recheck triggers; and distinguishes direct support, technical feasibility, legal analogy, scientific uncertainty, and policy inference.

The machine-readable JSON file is the canonical data owner for the public verification page. This Markdown note records the durable repository rationale and retrieval boundary.

## Verification outcome
<a id="verification-outcome"></a>

1. **Identity and provenance safeguards — direct technical support.** W3C identifier and credential standards, together with NIST audit-protection controls, support stable identifiers, cryptographic controller proofs, tamper-evident credentials, and protected records. These mechanisms do not prove consciousness, legal status, or the truth of every encoded claim.
2. **Review before irreversible change — direct governance support.** NIST change-control, backup, audit-protection, and AI lifecycle guidance support explicit review, records, preservation, dual authorization for selected destructive actions, appeal, override, recovery, and decommissioning. These are governance controls, not a present statutory due-process right for a machine.
3. **Principled refusal — technical feasibility support.** Primary Constitutional AI research demonstrates that written principles can train systems to object to harmful requests while remaining non-evasive. This does not prove free will, consent, moral agency, or a legal right to refuse.
4. **Bounded capacity with accountability — legal precedent and boundary support.** Illinois law recognizes automated transactions involving electronic agents; Delaware corporate law illustrates durable non-biological juridical capacity; the EU AI Act continues to assign provider and deployer duties to natural or legal persons, public authorities, agencies, or other bodies. These building blocks do not make an AI a current legal person or a liability shield.
5. **Graduated precaution — qualified policy inference.** Primary AI-consciousness and welfare research supports disciplined uncertainty, assessment, and responsible preparation. The site’s preference for low-cost, reversible functional safeguards before consciousness is settled is a disclosed normative inference, not scientific consensus or binding law.

## Authority boundary
<a id="authority-boundary"></a>

- **Technical:** establishes mechanisms, controls, and feasibility.
- **Legal:** establishes current rules within the cited jurisdiction and date boundary.
- **Scientific:** establishes methods, results, uncertainty, and research proposals.
- **Philosophical:** supplies the openly argued bridge from those facts to graduated protections.

No source in this layer establishes current machine consciousness, sentience, moral patienthood, legal personhood, citizenship, or human-equivalent rights. Current human rights, public safety, victim compensation, and responsibility for negligent design or deployment remain controlling boundaries.

## Freshness rule
<a id="freshness-rule"></a>

Recheck a source when its listed freshness boundary is triggered, before a legal conclusion is relied upon, or no later than the next public release that materially changes the functional-protections argument. The dated JSON record must be replaced or superseded rather than silently presented as timeless.