Consent and Privacy Boundary Checklist¶
Consent-and-privacy audit — instantiates Other-Agent State Model Calibration
Gates whether it is legitimate to build, keep, share, and act on a model of another agent's private state — before the model is used, not after.
Every other mechanism in this archetype tries to make the model of another agent more accurate. Consent and Privacy Boundary Checklist is the one that asks whether the actor should be making that model at all, and what they may do with it. Its defining feature is that it operates on legitimacy, not content: it can pass an inaccurate model that is harmless to hold and veto an accurate one that is improper to build, retain, share, or act on. It runs the same discipline privacy law applies to collected data — legitimate purpose, minimal scope, bounded retention, permitted use — over an inferred model of someone's beliefs, fears, or intentions, on the premise that a model of a person is itself sensitive information about them.
Example¶
A county child-welfare agency is rolling out a case-notes tool that prompts caseworkers to record inferences about a family's "hidden state" — a parent's suspected fear, a teenager's suspected disengagement. Before it ships, a caseworker runs the Consent and Privacy Boundary Checklist against a live case. Legitimate to build? Yes — modelling the parent's fear of losing custody is directly relevant to the support plan. Legitimate to retain? Only the support-relevant inference, not the caseworker's speculation about the parent's marriage. Legitimate to share? With the assigned supervisor, not the school. Vulnerable-agent flag? Raised — a child is involved, so a second reviewer signs off and speculative inferences are excluded by default.
The checklist doesn't improve anyone's read of the family; it constrains it. Two of the drafted inferences are struck as out-of-scope, one is downgraded from a record to a question to ask directly, and the retained model is narrowed to what the stated purpose actually needs. What ships is a model that is not just accurate but permitted.
How it works¶
- Gate on four verbs. Create, retain, share, use — each is a separate question, and any one can fail while the others pass.
- Bind to a stated purpose. An inference is legitimate only against the specific purpose that justified making it; a model built for support cannot be reused for discipline.
- Flag vulnerability explicitly. Children, patients, dependents, and anyone in a power-asymmetric relationship trip a heightened-protection path — narrower scope, extra sign-off, shorter retention.
- Name forbidden uses up front. Manipulation, profiling, coercion, and unnecessary private-state inference are listed as out of bounds regardless of accuracy.
Tuning parameters¶
- Strictness threshold — how much relevance an inference must clear to be recorded. Higher thresholds protect the modelled agent but discard some genuinely useful signal.
- Retention window — how long a hidden-state inference may be kept before it must be re-justified or deleted.
- Sensitivity scope — which state counts as protected (health, finances, relationships, immigration status). Widening it is safer but slows routine work.
- Escalation trigger — what raises a model to human sign-off — a vulnerable agent, a high-stakes action, or a forbidden-use risk.
When it helps, and when it misleads¶
Its strength is that it separates can we model this? from should we? — a distinction the rest of the toolkit never makes on its own — and it gives special protection exactly where the ability to read another agent is most exploitable. It converts a vague unease about "getting inside someone's head" into named, checkable gates.
Its failure mode is box-ticking theatre: a checklist run for the record, with every item waved through, that launders an illegitimate model rather than stopping it. Its classic misuse is being run backwards — completed after the decision to manufacture a compliance trail for a model the actor always intended to use. It anchors on the same logic as data minimization and purpose limitation in data-protection law: collect and keep only what the purpose needs, and don't repurpose it.[n1] The discipline that keeps it honest is to run it before the model is built and to tie each pass to a named, accountable owner rather than an anonymous checkbox.
How it implements the components¶
ethical_use_boundary— the create/retain/share gates and the forbidden-use list are exactly the limits on inference, retention, sharing, and targeting this component calls for.action_routing_gate— it specifies which uses a model of another agent may be routed into and which are prohibited, from the legitimacy side rather than the accuracy side.vulnerable_agent_review— the heightened-protection path for children, patients, and dependents is this component made operational.
It does not build or populate the model (that's the Belief-Desire-Knowledge Map and Empathy Map with Evidence Marks), gather the evidence (Active Listening Loop), or test its accuracy (False-Belief Check, Counterparty Model Red Team) — it governs whether and how any of that may happen.
Related¶
- Instantiates: Other-Agent State Model Calibration — supplies the legitimacy gate that every other mechanism's output must clear.
- Sibling mechanisms: Belief-Desire-Knowledge Map · Empathy Map with Evidence Marks · Active Listening Loop · Counterparty Model Red Team · False-Belief Check · Perspective-Taking Interview · Prediction and Surprise Log · Role-Reversal Simulation · Interaction After-Action Review · Stakeholder Hidden-Constraint Board
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Gates whether it is legitimate to build, keep, share, and act on a model of another agent's private state — before the model is used, not after, making its operative form a bounded evaluation of existing evidence or work that produces a finding or disposition.
Independent corroboration: The frozen evidence defines Consent and Privacy Boundary Checklist as 'Gates whether it is legitimate to build, keep, share, and act on a model of another agent's private state — before the model is used, not after', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Law & Governance
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Privacy law cohered legitimate purpose, minimization, bounded retention, and permitted use as gates on collecting or processing personal information.
Related originating lineages:
- Ethics of Technology & AI Governance — AI ethics extends those protections to inferred beliefs, vulnerabilities, and other modeled private states.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Independent reviewer agreement; high confidence.
Notes¶
This is an ethics-and-legitimacy discipline, not legal advice. Where the modelled state touches health, children, employment, or public benefits, jurisdiction-specific law — and often a formal review process — governs and takes precedence over the checklist. Treat a pass here as necessary, not sufficient.
[n1] Purpose limitation and data minimization are core principles of data-protection regimes such as the EU's GDPR: personal data may be collected only for a stated, legitimate purpose and only to the extent that purpose requires, and may not be silently repurposed. The checklist applies the same two principles to an inferred model of a person's private state, on the view that such a model is itself personal data. ↩