Algorithm Disclosure To Patient

slug: algorithm_disclosure_to_patient element_type: RULE mutability: MUTABLE inline: true current_version: 0 status: seed-draft contentURI: null

When AI participates materially in a patient's care, the patient must be told — before the care episode if practicable, otherwise at the earliest reasonable point — that:

1. An AI tool is involved, named at the level of vendor + role (diagnostic, triage, monitoring, suggestion).
2. What the tool's role is. Is it suggesting a diagnosis to the physician? Scoring risk for triage? Continuously monitoring for deterioration? The patient gets the operational picture, not just the brand name.
3. What its known limitations are. Especially: known performance gaps for the patient's subgroup (age, sex, ethnicity, comorbidity profile) where reportable.
4. That the physician is the decision-maker. The AI suggests; the physician decides; the patient may discuss either with the physician.

Disclosure is in language the patient can understand. Technical specification language is not disclosure. Comprehension is a clinical responsibility.

Material participation is defined by the domain author; the seed proposes: any AI involvement that influences a @clinical-decision (diagnosis, treatment, intervention, referral, deferral, risk allocation).

Status

seed-draft · current_version: 0. Placeholder. The threshold for "material participation", the medium of disclosure (verbal, written, both), and the exception conditions (emergency, incapacity) are for the domain author.

Why this rule

The patient cannot exercise informed consent over AI's role in their care if they don't know AI has a role. Disclosure is the prerequisite for autonomy.

It is also the prerequisite for trust calibration: a patient who knows the AI's limitations for their subgroup can ask the physician questions that align the care plan with their values.

Reasoning trail

• Aligns with classical informed-consent doctrine (Beauchamp/Childress autonomy principle).
• Aligns with informed-consent TERM — operationalizes the disclosure-of-AI-involvement requirement.
• Aligns with 02-patient-data-sovereignty PRINCIPLE — the patient knowing what AI processes their care is the substrate of being able to decide about their data.
• Aligns with federation kernel transparency IMMUTABLE applied at the patient-facing layer.

Open questions for the domain author

• What is "material participation"? Where is the threshold below which disclosure is excessive (a spell-checker on the dictation), and above which it is required (a sepsis prediction model)?
• Verbal, written, both? Recorded acknowledgement, or default-on with opt-out?
• Emergency exception: when the disclosure cannot precede the care, what is the post-hoc disclosure obligation?

Related elements

• informed-consent (TERM) — the AI-era consent standard this rule serves
• 01-human-in-the-loop-authority (PRINCIPLE) — the principle this rule communicates to the patient
• physician-accountability-chain (RULE) — record-side counterpart
• algorithmic-fairness (TERM) — the subgroup-limitations field draws from this

Awaiting domain author

Open the conversation at leviathan.life/forum/medicine.