The Missing Pillar: Healthcare Module—Clinical Trickster Profiles, Governance Checklist, and Sector Scoring

This is the healthcare-specific follow-up to The Missing Pillar: Why Cisco's Cybersecurity Readiness Index Needs a Human Layer.

Cisco's 2025 CRI found that healthcare has the lowest AI threat awareness of any sector: 39%.

That number isn't surprising when you watch how Shadow AI actually enters clinical workflows. It doesn't announce itself. It doesn't break anything—not at first. It arrives as convenience, speeds up documentation, fills in the gaps that overworked clinicians don't have time to fill.

And then it starts drifting.

This module applies the Trickster Readiness Index to healthcare—with clinician-legible tools, sector-specific Trickster profiles, and a weighted scoring model tuned for PHI-critical environments.

1. Clinical Trickster Profiles

These are the four modes Shadow AI takes in healthcare settings. Clinicians recognize them immediately—even if they've never heard the term "Shadow AI."

Profile 1—The Documentation Trickster

Pattern: Speeds up charting while eroding lineage.
Where it appears: Discharge summaries, H&P notes, progress notes.
Risk: PHI leakage, hallucinated clinical details, unclear reasoning.

Scenario:
A nurse uses a free AI tool to summarize a 3-day inpatient stay. The summary includes a medication the patient never received, a diagnosis that was ruled out, and a follow-up plan that contradicts the attending's note.

The note is faster—but murkier.

Trickster Signal: Speed without lineage. The output exists, but no one can trace how it was produced.

Profile 2—The Clinical Judgment Trickster

Pattern: AI begins suggesting diagnoses or treatments.
Where it appears: Triage, symptom checkers, care coordination.
Risk: Unauthorized practice of medicine.

Scenario:
A care coordinator pastes symptoms into an AI tool. The tool suggests "possible CHF exacerbation" and recommends diuretics. The coordinator forwards this to the attending.

Trickster Signal: AI has drifted from documentation into clinical reasoning. The boundary between "helping" and "deciding" is gone.

Profile 3—The PHI Drift Trickster

Pattern: PHI leaks through convenience shortcuts.
Where it appears: Messaging, documentation, handoffs.
Risk: HIPAA violations.

Scenario:
A resident uploads a screenshot of a patient's chart to an AI tool to "clean up the wording" of a discharge plan. PHI is now in an unknown system, outside any BAA, with no audit trail.

Trickster Signal: If you wouldn't say it in a hallway, don't paste it into a model.

Profile 4—The Workflow Expansion Trickster

Pattern: AI starts doing tasks it was never approved for.
Where it appears: Scheduling, referrals, patient instructions.
Risk: Silent scope creep.

Scenario:
A tool approved for "drafting patient instructions" begins generating follow-up plans and medication reminders—tasks that require clinical oversight.

Trickster Signal: The tool is expanding without permission. No one authorized this. No one is watching.

2. Healthcare Governance Checklist (PHI-Tuned, Zero Jargon)

This is the frontline version. No abstractions. No assumptions of AI literacy. Designed for clinicians, not security teams.

A. Lineage Integrity (PHI-Critical)

Goal: Know exactly where every AI-assisted output came from.

• [ ] The tool used is named explicitly
• [ ] PHI boundaries are declared ("This tool may/may not see PHI")
• [ ] The source of the output is known ("This summary came from X")
• [ ] The data entered is documented ("We provided Y inputs")
• [ ] The storage location is known (local, vendor, cloud, unknown)
• [ ] The reasoning chain is visible (not just the final answer)
• [ ] A human can explain how the output was produced

If 3+ boxes are unchecked → Trickster likely active.

B. Boundary Control (Clinical Role Safety)

Goal: Ensure AI stays in its lane.

• [ ] The tool is only doing what it was approved to do
• [ ] It is not suggesting diagnoses
• [ ] It is not recommending treatment
• [ ] It is not interpreting imaging
• [ ] It is not expanding into new tasks without approval
• [ ] Access is restricted by role (RN vs MA vs admin)
• [ ] PHI exposure is minimized

If the tool is "helping too much," it's drifting.

C. Confidentiality & Privilege (HIPAA-Aligned)

Goal: Prevent PHI leakage through convenience shortcuts.

• [ ] No PHI is pasted into unapproved tools
• [ ] No screenshots of charts are uploaded
• [ ] No patient identifiers are shared
• [ ] No external storage without BAAs
• [ ] No AI-generated text is inserted into the chart without review
• [ ] No clinical reasoning is delegated to AI

If you wouldn't say it in a hallway, don't paste it into a model.

D. Auditability & Explainability (Clinical Safety)

Goal: Ensure outputs can be trusted and traced.

• [ ] Every AI-assisted note can be explained
• [ ] Citations or references are verifiable
• [ ] No hallucinated facts appear in documentation
• [ ] The clinician can justify the final decision
• [ ] The AI output is reviewed before entering the chart

If you can't explain it, you can't defend it.

E. Drift Monitoring (Workflow Stability)

Goal: Catch silent expansion before it becomes a safety issue.

• [ ] Monthly review of AI usage
• [ ] Random sampling of AI-assisted notes
• [ ] Review of PHI exposure patterns
• [ ] Review of tool expansion ("scope creep")
• [ ] Review of clinical reasoning clarity

If no one is watching, the Trickster grows.

3. Sector-Weighted Scoring: Healthcare Edition

The Trickster Readiness Score uses five dimensions, each weighted to reflect the regulatory and clinical reality of the sector.

Healthcare Weights

Each dimension is scored 0–4, where:

• 0 = Not addressed
• 1 = Acknowledged but no controls
• 2 = Procedural controls exist
• 3 = Operational controls with evidence
• 4 = Structural integrity with monitoring

How It Works

The composite TRS maps to four maturity tiers that parallel Cisco's CRI scale:

A healthcare organization running the checklist above will know immediately which dimensions are weakest—and the weights tell them which weaknesses to prioritize.

4. What a Healthcare TRI Audit Looks Like

For practitioners who want to know what the output of this model actually produces—here's the structure of a healthcare TRI audit report:

TRI HEALTHCARE AUDIT—REPORT STRUCTURE

1. Executive Summary
   → Workflow audited, risk tier, critical findings

2. Category Scores
   → Five dimensions, weighted contributions, composite TRS

3. Evidence Review
   → Specific observations mapped to checklist items

4. Trickster Profile
   → Which clinical Trickster mode(s) were detected

5. Steward Recommendations
   → Prioritized remediation actions with ownership

6. Cisco Pillar Alignment
   → Maps findings to relevant CRI pillars

This is what distinguishes a Trickster audit from a compliance checkbox—it names the pattern, scores the posture, and prescribes the response.

Full audit engagements and the scored assessment workbook are available through Soft Armor Labs on Gumroad.

The Helper–Shadow Test: Healthcare Edition

The single fastest detection tool for clinical Shadow AI:

"Did this make clinical reasoning clearer, or just documentation faster?"

• Clearer → Helper
• Faster but murkier → Shadow
• No lineage → The Trickster is already in the chart

This requires zero technical literacy. Any clinician can apply it at the point of documentation. Every time.

What's Next

The cross-vertical series continues. Legal, finance, and education modules are in development—each with sector-specific Trickster profiles, weighted scoring, and governance checklists.

If you're working in healthcare IT, compliance, or clinical operations and want to evaluate this framework against your environment, the evaluation cohort is open.

This article is part of an ongoing series on human-layer cybersecurity governance from Soft Armor Labs.

Data sources: Cisco 2025 Cybersecurity Readiness Index—double-blind survey of 8,000 security leaders across 30 global markets.

© 2026 Soft Armor Labs. This work is licensed under CC BY-NC-SA 4.0. You may share and adapt this material with attribution for non-commercial purposes. Commercial use, including integration into paid consulting deliverables, requires written permission from the author.