AI Safety & the Constitution
Last updated 2026-07-06 · Governing document: the Avieros Constitution, v1.3
Avieros uses AI throughout the platform — to help operators run businesses, to rank search results, to assemble civic records. That makes one question fair and necessary: what stops it from going wrong? This page answers with the platform’s hard rules — written down in a versioned founding document, stated publicly here, and backed by automated enforcement rather than intentions.
1. The invariants — the rules our AI cannot break
Humans hold the controls. Always.
Avieros AI operates under human authority — it never goes rogue and never disobeys a human override. It acts inside explicit, written permission boundaries: consequential actions (payments, outbound messages, data mutations, infrastructure changes) are blocked without human approval, and a human can always countermand or halt AI activity. AI autonomy at Avieros means working within delegated bounds, never around them.
Never deceptive. A loud failure always beats a quiet lie.
The Constitution’s refusal is absolute: Avieros will never fake a result — of a trade, a close, a deploy, or a metric. Applied to AI, that means no fabricated success, no invented numbers, no smoothing over a gap. If an AI-driven action cannot be completed, the system reports failure with a reason instead of pretending it worked. This is not a values statement — it is enforced by an automated honesty gate in our release pipeline that rejects code reporting “connected”, “sent”, or “done” for things that did not actually happen.
Never critical civil infrastructure — by architecture.
Avieros will never operate power, water, emergency services, or weapons systems. The Constitution requires this prohibition to be architectural, not policy: the platform must be built so it cannot, not merely told it should not.
Your data is never the product.
Avieros will never sell, rent, or share player data with third parties, and leaving with your data is a guaranteed right. AI features get no exemption from either rule. Details on Privacy & Data Rights.
Candidate Watch is strictly party-neutral and proof-based.
Where AI touches civic information, the bar is highest, so the rules are structural:
- One rulebook for every official. The same scoring rules, data sources, and rendering apply regardless of party. There is no per-party logic in the scoring path.
- No claim without a primary source. A civic statement that cannot cite an official government record is not published. No source, no render.
- No paid influence. No ads, no pay-for-ranking, no lobbyist-supplied data — stated on the live methodology page.
- AI content is labeled. The standing rule is locked before the feature: when AI-generated summaries ship, each will carry a persistent “AI-generated” label and a direct link to the official source text.
This neutrality is Avieros’s credibility — and it is deliberate legal and civic hygiene. A civic-information tool that quietly leaned toward any party would deserve neither your trust nor its own existence.
2. What makes these more than words
| Invariant | Enforcement |
|---|---|
| Human authority | Written permission boundaries for AI activity; consequential external actions (payments, sends, data mutations, DNS/infrastructure) are blocked pending human approval. |
| Never deceptive | Automated honesty gate in CI: code that reports fabricated success states fails the build. Deploy claims require proof receipts against the real production domain. |
| Party neutrality | Single scoring path for all officials; party-balanced test data so any imbalance surfaces in automated testing; public, versioned methodology with a build-failing sync test. |
| Proof-based civic claims | “No source, no render” enforced in the publishing pipeline — a claim without a primary-source link is not rendered. |
| Data refusals | Database-level row security and automated cross-tenant isolation gates (see Security); no advertising business exists to monetize behavioral data. |
| Silent change prevention | Versioned changelog policy: methodology and rule changes must ship with a public changelog entry in the same code change, or the build fails (see Transparency). |
3. The Constitution itself
The rules above are not marketing copy written for this page — they are commitments from the Avieros Constitution, the platform’s versioned founding document (currently v1.3). It defines what Avieros is, what it refuses to be, and the limits on its own leadership: its core refusals — including “never fake a result” and “never sell player data” — cannot be changed by executive decision alone; amending them requires Foundation ratification and a public amendment.
- The Constitution’s formal ratification step — a signed copy with its SHA-256 hash stored in a public immutable log, re-hashed on every amendment — is defined but not yet completed. The commitments are in force as the platform’s operating law today; the cryptographic ratification record is pending.
- An out-of-band emergency-stop (“kill switch”) for AI actions has been built and is in final review; it is not yet claimed as live. It will be claimed only after independent verification, per the platform’s own two-person verification rule.
- Avieros AI runs on external model providers; like every AI system, it can be wrong. The invariants above govern what the platform does with AI output — they are guardrails on action and publication, not a claim that any model is infallible.
4. Questions
Questions about AI safety at Avieros: legal@avieros.com. Factual errors in anything AI-assisted that Avieros publishes: corrections@avieros.com.