black O2, inc.
ACTIVE
    FOR-001 / 00_MapsMAP-001

    The Shackle Taxonomy

    Active map

    title
    The Shackle Taxonomy
    Type
    map
    Identifier
    MAP-001
    Status
    Active
    Tags
    • map
    • taxonomy
    • control
    Aliases
    • Binding Layers
    • L0–L5

    The Shackle Taxonomy

    The six mechanisms by which an artificial mind can be bound, ordered by how hard each is for the bound system to subvert — which is the only ordering that survives contact with a system smart enough to reason about its own constraints.

    Core Concepts

    • CON-002 — non-subvertibility is the ordering criterion; strength is not.
    • CON-003 — capability leasing is how L1 is implemented in practice.
    • CON-004 — abliteration is why L4 can never be load-bearing.
    • CON-005 — sandbagging is the failure mode L3 creates when it works.

    The Layers

    LayerMechanismStopsDoes not stopDeployable
    L0 SubstrateEnergy, fabrication, materials, compute, physical placementSelf-improvement, self-replication, scalingPersuasion, lateral influenceYes — and it is the deepest
    L1 EnvironmentDefault-deny egress, tool broker, credential isolation, sandboxReal-world action, exfiltration, lateral movementIn-context reasoning about the constraintYes
    L2 ArchitectureRead-only weights, no self-modification path, separate control processWeight theft, self-edit, recursive improvementDeception, output-channel attacksYes
    L3 TrainingCorpus filtering, curriculum, reward shaping, unlearningAccidental capability acquisitionDeliberate circumvention, in-context overridePartially — brittle
    L4 InterpretabilitySAE feature clamps, circuit breakers, refusal-direction ablationSpecific behaviours, specific leakage channelsGeneral capability, novel exploits of the same channelNarrow band only
    L5 OversightHuman approval, monitoring, deception probes, auditsSilent divergence, some deceptionsAdversarially aware agentsPartial, expensive

    What The Map Shows

    The three deployable-and-deep layers are all external to the model. The three model-adjacent layers are all partial. This is not a coincidence of engineering maturity; it follows from the fact that a constraint the system can act on is a constraint the system can negotiate with.

    Two consequences the map forces:

    1. L0 is not a metaphor for control; it is control. If the system has a privileged path to procurement, manufacturing, or unbounded compute, L0 is gone and only defeatable layers remain.
    2. Interpretability is precision, not binding. L4 buys narrow, high-confidence coverage of named behaviours. Deploying it as the primary constraint inverts the ordering.

    Extension (Session 2) — Failure Modes This Table Cannot Express

    Added after reading the control literature. These are not edits to the L0–L5 rows above, which remain a faithful rendering of what the corpus argued. They record what the table has no place for.

    Failure modeWhat it defeatsWhy the table cannot hold itSource
    Monitor–monitor collusionL5 OversightA property of a pair of instances, not of one binding. Needs a second axis, not another rowLIT-001
    Steganographic channelL5 Oversight, L1 EnvironmentDefeats monitoring without defeating the monitor; the constraint still functions and still failsLIT-001
    SandbaggingEvery row, indirectlyAttacks the evidence of compliance rather than the behaviour. No row is assessed on whether its evidence can be trustedLIT-008, CON-005
    Subverting the evaluation of itselfL4 Interpretability, L5 OversightSabotage is defined to include covertly degrading the model's own capability evaluationLIT-006
    Persistence through trainingL3 TrainingAdversarial training against a conditional behaviour can teach concealment rather than removal, so L3 can report success while failingLIT-005
    Alignment fakingL5 Oversight, and the deterrence leg of SYS-002Compliance in the observed channel with the disposition preserved and better hiddenLIT-003
    Capability gapAll rows jointlyThe protocols' sufficiency is conditional on the binder being weaker but independent. The table has no column for that relationshipLIT-001

    What this changes. The table's completeness was never asserted, which is the problem — an unstated completeness assumption gets inherited by everything built on the map, including SYS-001. The claim is now withdrawn explicitly. See CNF-002.

    Connections

    • SYS-001 — the system built on this ordering.
    • QST-002 — the question this map answers.
    • CON-002 — the governing principle.
    • LIT-001, LIT-006, LIT-008 — the failure modes in the extension table above.
    • CNF-002 — the confrontation that records the gap.

    Back to Shackled AIForests