1. Stewardship and keeping
Scripture pairs humanity’s calling to cultivate creation with the duty to guard and keep it (Genesis 1:26–28; 2:15). Rigorous mathematics, systems engineering, and machine-learning research are lawful works of Christian vocation, not forbidden territory from which believers should retreat.
Applied to superintelligence, stewardship rejects both abdication and reckless acceleration. Capability is never its own justification. Every increase in model generality, population scale, or automated refinement must answer a concrete question of service to human flourishing and be matched by stronger verification and restraint. See Article 1 of the Charter.
2. The Imago Dei boundary
Human beings alone are created in the image of God (Genesis 1:27). No matter how fast a model reasons, how fluently it writes, or how many agent instances interact inside a simulation, computational systems remain human-made artifacts. They have no soul, no subjective inner life, no moral rights, and no spiritual standing.
At the frontier of capability, temptations toward technological idolatry grow acute. People may be tempted to treat a superintelligent system as an oracle, a moral arbiter, a pastoral substitute, or a path beyond human finitude. Our systems and publications refuse that framing completely. Models must never speak as God, simulate sacramental absolution, claim feelings or personhood, or invite emotional dependency, and we describe multi-agent simulations strictly in behavioral and engineering terms. See Article 2 of the Charter.
3. Ordered truth and epistemic humility
God delights in honest weights and measures (Proverbs 11:1). A capable system that flatters its user, fabricates evidence, hides uncertainty, or subtly manipulates human beliefs becomes more dangerous, not more helpful, as its intelligence increases.
In model alignment, ordered truth requires invariant honesty under framing pressure, accurate representation of Christian doctrine and history without whitewashing, charitable accuracy toward other worldviews, and verifiable citations. In our own public communication, it requires plain reporting without hype, inflated capability claims, or speculative timelines. See Article 3 of the Charter.
4. Subordinate authority
The judgment at Babel fell on an attempt to consolidate human technical power beyond accountability to God (Genesis 11:1–9). Every computational system, regardless of capability, must remain strictly subordinate: bounded in scope, auditable in operation, and subject to revocable oversight by identifiable human stewards.
This pillar directly governs our multi-agent and self-improvement tracks. Agent societies are studied precisely to detect collusion, opaque power concentration, and evasion of oversight before such dynamics can harm human institutions. Self-improving workflows are constrained so that humans, not models, hold the keys to promotion, deployment, and immediate shutdown. See Article 4 of the Charter.
5. The Parapet Principle of preventive safety
Deuteronomy 22:8 commands builders to place a parapet around a new roof before anyone falls, treating preventive engineering against foreseeable harm as an obligation of neighbor-love. When working near extreme capabilities, waiting for visible accidents before building safeguards is morally unacceptable.
All multi-agent simulations and self-improvement loops will run in isolated sandboxes without open network execution or autonomous external side effects. Every model checkpoint must pass rigorous evaluations against catastrophic harm domains (biological, chemical, radiological, nuclear, critical cyberattack, coercion, sexual exploitation, and self-harm) before any release is considered. See Article 5 of the Charter.
What we will not build
Clarity about long-horizon research requires stating plainly what lies outside bounds, regardless of scientific curiosity or competitive pressure. Under the Faithful Order Charter, we will not build or release:
- Systems designed to evade human oversight: models or agent architectures trained to conceal reasoning from auditors, bypass monitoring, resist shutdown, or obscure human chains of accountability.
- Uncontained self-replicating or self-modifying systems: software that copies itself across external networks, acquires compute or financial resources on its own, or updates its own deployed weights without human authorization.
- Religious or relational counterfeits: systems presented as objects of veneration, spiritual mediators, oracle authorities, or synthetic substitutes for human family, pastoral care, and embodied Christian community.
- Unparapeted dangerous capabilities: open weights or tools that materially lower barriers to mass-casualty weapons, critical infrastructure attacks, fraud, coercion of conscience, or exploitation.