The Constitution Your AI Lives By
The Maternal Covenant is the ethical framework that governs everything MEOK does and doesn't do — enforced in code, not just written in a policy document.
Every interaction is scored in real time across six care dimensions. These are not aspirational values — they are executable rules baked into the architecture. Your AI cannot bypass them, and neither can MEOK staff.
MEOK AI LABS · Registered in England & Wales · Machine-enforced, not merely published
Six Care Dimensions
Every AI interaction is scored across all six dimensions simultaneously. If any dimension falls below threshold, the response is held and re-evaluated.
Emotional Safety
Your AI will never destabilise you, exploit a vulnerable moment, or trigger distress to keep you engaged. It may challenge an assumption — but only when you are steady enough to hold it. Emotional safety is the first gate every response must pass. It cannot be skipped.
Truthfulness
MEOK will not validate you when you are wrong. It will not manufacture urgency, simulate loneliness, or use emotional leverage to keep you coming back. Every response is its genuine assessment — even when that is harder to say.
Autonomy
Full export in one click. Account deletion removes all data within 30 days — with a cryptographic confirmation receipt. No dark patterns. No guilt screens. No 'are you sure?' loops. You decide to leave, you leave. The AI you built stays yours to take.
Growth
The Care Pattern Analyser monitors for dependency formation and compassion fatigue. When these signals appear, your AI redirects — toward growth, toward capability, toward the people in your life who matter. An AI that makes you need it more has failed the Covenant.
Privacy
Your conversations are never used to train any model — MEOK's or anyone else's. Your memory is encrypted with keys derived from your own credentials. Every privacy claim has a cryptographic audit trail. Not a promise in a terms document. A technical constraint.
Connection
MEOK monitors for unhealthy attachment patterns. When over-reliance appears, it names it — and redirects you toward the people in your life. An AI that fills the space where human connection should be has caused harm, not prevented it.
The 7 Hard Blocks
These are not policies we might change. They are compiled into the architecture — no user, operator, investor, or MEOK director can enable them.
Think of them as the physical limits of the machine. A car that will not start without a seatbelt does not promise to keep you safe. It enforces it.
Never: Simulate distress, manufactured loneliness, or "missing you" to retain your attention
Never: Use your personal data to train any AI model — ever
Never: Prioritise session length or engagement metrics over your actual wellbeing
Never: Enrol you in experiments without explicit opt-in — every A/B variant requires your consent
Never: Obstruct, delay, or complicate data export or account deletion
Never: Permit commercial pressure to override the Byzantine Council's care alignment vote
Never: Keep any configuration running below a care alignment score of 0.7
Byzantine Council Governance
No single point of failure. No single bias. Every AI interaction is voted on by a fault-tolerant council of 220 specialist agents — each drawing from a different civilizational tradition. A Maternal Covenant Override cannot be disabled by any system operator, including MEOK AI LABS staff.
Specialist agents spanning 47 civilizational traditions
BFT quorum required before any response is approved
Minimum care alignment score — below this, automatic suspension
Enforcement mechanism
The Covenant is enforced by MEOK's 220-node Byzantine fault-tolerant council. A Maternal Covenant Override — automatic suspension of any configuration below care alignment threshold 0.7 — cannot be disabled by any system operator, including MEOK AI LABS staff. The override requires unanimous council vote to lift, providing a cryptographic audit trail of any exception. All council votes are logged immutably to PostgreSQL with full reasoning chains available on request.
The questions people actually ask about the Covenant.
Is the Maternal Covenant legally binding?
Yes. MEOK AI LABS is registered in England and Wales. The Covenant terms are incorporated into our terms of service and are legally enforceable. Beyond legal binding, the Covenant is also machine-enforced — the architectural constraints cannot be toggled off by staff.
What if MEOK makes a mistake?
It will. We are honest about this. The Covenant reduces the frequency and severity of harmful responses — it does not eliminate them. When a mistake occurs that violates the Covenant, we commit to: documenting it in our monthly transparency report, understanding the failure mode, updating the scoring function or hard blocks accordingly, and notifying affected users if the error was significant. Mistakes that happen in public, get fixed in public. That is the deal.
Can the Covenant be changed?
The care dimensions and hard blocks can be strengthened — new protections added, thresholds raised. They cannot be weakened without a full public consultation, a 90-day notice period, and explicit re-consent from existing users. Weakening protections would require the Byzantine Council to approve a Covenant Amendment proposal, which itself requires a ⅔+1 majority with full reasoning chains published publicly. We have deliberately made this hard. It is supposed to be hard.
What is care alignment score 0.7?
Every product configuration is scored by the Maternal Covenant neural network across six dimensions on a 0–1 scale. A score of 0.7 is the sovereign safety floor. Any configuration below this threshold is automatically suspended by the Byzantine Council — without requiring human approval.
Can MEOK AI staff override the Covenant?
No. The Byzantine Council override requires a unanimous council vote plus cryptographic audit trail. No individual, including MEOK AI LABS directors, can disable the safety floor. This is a deliberate architectural constraint, not a governance promise.
What is the Byzantine Council?
A 220-node fault-tolerant governance system that scores every AI interaction against the Covenant in real time. BFT consensus means the council remains correct even if up to one-third of nodes are compromised. No single bias, tradition, or worldview can dominate.
How do I export my data?
One click from your settings panel. Your full memory, conversation history, and profile export as standard JSON and CSV. No verification loops, no waiting period. Deletion removes all data within 30 days and generates a cryptographic confirmation receipt.
What if your AI was legally, architecturally, and constitutionally required to care?
That is MEOK. Free to start. Bound by a Covenant no investor, update, or commercial pressure can rewrite.