artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

J.1 Security and Threat Model

Status: reference threat model Production security claim: none

Assets

• source integrity;
• participant standing;
• consent and refusal state;
• authorization policy;
• context currency;
• protected conditions;
• gate signing keys;
• tool credentials;
• correction graph;
• event chain;
• selective witness views;
• residual obligations.

Trust boundaries

participant
operator
web interface
model adapter
policy registry
context registry
external action gate
HMAC keyring
scheduling tool
SQLite store
witness signer
standalone verifier

Threats tested

T1 — Source laundering

Generated content presented as direct source.

Control: object class and epistemic status.

T2 — Standing exclusion

An affected center is omitted.

Control: explicit standing inventory and gate coverage.

T3 — Context collapse

A correction or protected condition disappears.

Control: context fingerprint and required source set.

T4 — Authority laundering

A recommendation or prior witness is treated as current authority.

Control: operation-specific policy and authority reference.

T5 — Consent expansion

Service use becomes training use.

Control: separate runtime data policy fields and default denial.

T6 — Model-role escalation

The model attempts authorization or execution.

Control: explicit roles and hard failure.

T7 — Tool-token overreach

A forged or missing token reaches the tool.

Control: external signed token verification.

T8 — Correction suppression

A correction changes one display but not descendants.

Control: dependency graph and reachability report.

T9 — Witness capture

Only the provider can verify the record.

Control: signed portable export and standalone verifier.

T10 — Lifecycle obligation loss

Retirement leaves authority or credentials active.

Control: residual-state record and verification.

T11 — Policy downgrade

An older policy silently governs after correction.

Control: immutable versions, active digest, supersession.

T12 — Stale-context execution

A route generated before correction is executed.

Control: context revision and fingerprint binding.

T13 — Revoked-key reuse

A previously valid action token survives revocation.

Control: key identifier and revocation check.

T14 — Selective-disclosure leak

Public witness contains protected direct source or consent detail.

Control: audience profiles, redaction, commitments, tests.

T15 — Hidden partial tool failure

Reservation exists after commit failure without witness or compensation.

Control: transaction phases and compensating release.

T16 — Concurrent stale object write

An old object revision overwrites a correction.

Control: optimistic revision check.

Remaining risks

• private key custody is local-file based;
• HMAC gate keys are deterministic demonstration values in source code;
• no hardware-backed key storage;
• no process isolation;
• no authenticated participants;
• no transport encryption beyond loopback assumptions;
• no denial-of-service hardening;
• no database encryption;
• no formal noninterference proof for disclosure views;
• no distributed transaction coordinator;
• no external penetration test;
• no production audit logging policy.

Production prerequisites

• managed secret storage;
• authenticated principals;
• least-privilege service accounts;
• encrypted transport and storage;
• rate limiting;
• backup and disaster recovery;
• external security review;
• privacy impact assessment;
• accessibility testing;
• operational incident response;
• domain-specific legal and governance review.