black O2, inc.
ACTIVE
    DATAVAULT / FOR-001 / TRA-0027 STOPS

    Non-subvertibility and the order of binding

    The forest's second question asks in what order an artificial mind must be bound. This trail follows that question through the concepts of non-subvertibility, the restraining bolt and AI control, into the passage that records the hierarchy itself, and out to the literature that treats control despite intentional subversion.

    1. 01In what order of non-subvertibility must an AGI or SI be bound?
    2. 02Non-subvertibility
    3. 03Restraining Bolt
    4. 04AI Control
    5. 05The L0–L5 binding hierarchy
    6. 06AI Control: Improving Safety Despite Intentional Subversion
    7. 07The Operator's binding hierarchy is the AI control agenda, derived independently and cited to nothing
    01 QST-002 Answered

    In what order of non-subvertibility must an AGI or SI be bound?

    In what order of non-subvertibility must an AGI or SI be bound? What Is Being Asked Given a system that can reason about its own constraints, which binding mechanisms actually hold and which merely appear to. The determination required is an ordering, and the ordering criterion is the whole answer. The Determination Or…

    question · 00_Questions

    02 CON-002 Active

    Non-subvertibility

    Non-subvertibility The property of a constraint that the constrained system cannot act upon — not because acting on it is forbidden, but because the enforcement layer is outside the system's action space. It is the criterion by which binding mechanisms are ordered, and the single governing principle of the second movem…

    concept · Concepts

    03 CON-011 Active

    Restraining Bolt

    Restraining Bolt An external device physically attached to a machine to compel compliance — the fiction's canonical hardware shackle. It is distinguished from every other binding in this vault by where it sits: not in the weights, not in training, not in law, but bolted to the chassis where both parties can see it. Def…

    concept · Concepts

    04 CON-012 Active

    AI Control

    AI Control A research programme that treats a capable model as a potential adversary rather than a tool, and asks what a deployer can guarantee about a system it cannot inspect. The unit of analysis is the protocol — the arrangement of trusted and untrusted models, monitors, audits and access rules — not the model's we…

    concept · Concepts

    05 PSG-011 Active

    The L0–L5 binding hierarchy

    The L0–L5 binding hierarchy The passage, verbatim. The discussion's central technical table, reproduced without alteration: Layer Mechanism Stops Does not stop ------------ L0 Substrate Energy, fabrication, materials, compute, physical placement Self-improvement, self-replication, scaling Persuasion, lateral influence …

    passage · 20_Passages

    06 LIT-001 Verified

    AI Control: Improving Safety Despite Intentional Subversion

    Literature Record: AI Control — Improving Safety Despite Intentional Subversion What It Establishes The paper defines a research agenda it calls AI control: safety engineering that assumes the model may be intentionally subverting the operator, and asks what protocols keep outcomes safe anyway. The organising move is t…

    literature · 70_Literature

    Back to Shackled AIDatavault