Appearance
Public Figure Reconciliation
GhostAI's public figure reconciliation prevents well-known names from being masked as PII, so questions about public and historical figures get accurate AI responses.
The Problem
Standard PII detection treats all person names equally. When a user asks:
Who is Albert Einstein?
Without reconciliation, this becomes:
Who is [PERSON_1]?
The AI has no idea who [PERSON_1] refers to and cannot provide a useful answer. Public figure reconciliation solves this by recognizing that "Albert Einstein" is a public figure whose name does not need protection.
How It Works
When a PERSON entity is detected during PII scanning, GhostAI evaluates whether the name belongs to a recognized public or historical figure. If it does — and the name passes safety validation — it is preserved in the prompt rather than tokenized.
The system uses multiple resolution tiers to verify public figures, combining curated reference data with external knowledge sources. Each tier can be independently enabled or disabled by administrators.
Safety Filters
Before any name is reconciled, GhostAI applies safety filters to prevent private individuals from being misidentified as public figures. These filters analyze contextual signals — such as the presence of other PII near a name — and cross-reference detection confidence to ensure that only genuinely public names pass through.
This defense-in-depth approach means that even if a private individual shares a name with a public figure, the system errs on the side of privacy protection.
Resolution Tiers
GhostAI resolves public figures through multiple tiers, each with different characteristics:
| Tier | Description |
|---|---|
| Curated Lists | A maintained collection of widely recognized public and historical figures across categories including political leaders, scientists, authors, artists, athletes, and more. Provides fast, offline resolution with high confidence. |
| External Verification | For names not found in curated lists, GhostAI queries external knowledge sources to determine public figure status. Results are cached to minimize latency on repeated lookups. |
| Model-Based Fallback | When other tiers cannot resolve a name, GhostAI can use its configured language model to assess whether the entity is a public figure. This tier applies stricter confidence requirements to prevent false positives. |
Example
User types:
My colleague Sarah Johnson (sarah.johnson@acme-corp.com) asked about Albert Einstein.
What the AI receives:
My colleague [PERSON_a1b2] ([EMAIL_c3d4]) asked about Albert Einstein.
"Albert Einstein" passes through untouched because it is a recognized public figure with no correlated private PII. "Sarah Johnson" is tokenized because contextual signals indicate she is a private individual.
What Gets Reconciled vs. What Doesn't
| Input | Outcome | Why |
|---|---|---|
| "Tell me about Marie Curie" | Preserved | Recognized public figure, no private PII context |
| "Email john.smith@company.com about the meeting with John Smith" | Tokenized | Name correlates with a private email address |
| "Compare leadership styles of Abraham Lincoln and my manager David Chen" | Lincoln preserved, Chen tokenized | Lincoln is a public figure; Chen appears in a private context |
Enabling Public Figure Reconciliation
Via Settings UI
- Navigate to Settings → PII Detection
- Locate the Public Figure Reconciliation toggle
- Enable the feature
- (Optional) Configure individual resolution tiers and sensitivity
Permission Model
Public figure reconciliation follows GhostAI's standard permission hierarchy:
| Role | Capabilities |
|---|---|
| Org Owner / Admin | Set organization-wide default (enabled/disabled), lock setting to prevent user overrides |
| Team Admin | Set team-level default if org allows overrides |
| User | Toggle on/off for own sessions if org/team settings permit overrides |
When an admin locks the setting, users see the toggle in a disabled state with a tooltip explaining the policy.
How Reconciled Entities Appear in the UI
When a name is reconciled as a public figure:
- The name is not masked — it appears as the user typed it
- A green "Public Figure" badge appears next to the entity in the detection panel
- The entity is excluded from the PII count in the message badge
API Endpoints
| Method | Path | Description |
|---|---|---|
POST | /api/detect/public-figure | Check whether a name is a recognized public figure |
Compliance Considerations
- Privacy-first default: When a name is ambiguous, the system defaults to tokenizing it for safety
- Auditability: All reconciliation decisions are logged and available for review
- Admin control: Organizations can enforce policies at the org or team level, including disabling the feature entirely
- No PII leakage: Public figure reconciliation only affects names that pass all safety filters — private individuals are never exposed
Best Practices
- Start with curated lists only — enable additional resolution tiers as needed for your use case
- Review reconciliation logs — check for false positives in your domain
- Use context-rich prompts — surrounding text helps disambiguate names
- Keep sensitivity settings appropriate — higher thresholds reduce false positives at the cost of more public figures being tokenized
Troubleshooting
Public Figure Still Being Masked
- Verify the feature is enabled in Settings → PII Detection
- Check that the relevant resolution tier is active
- Ensure the name is a full match (partial names may not be reconciled)
Private Names Being Passed Through
- Verify contextual PII signals are present near the private individual's name
- Increase the sensitivity threshold for resolution tiers
- Report false positives via Settings → PII Detection → Feedback
Next Steps
- PII Detection — Core detection engine
- Tokenization — How non-public PII is protected
- Anti-Exfiltration — Obfuscation detection
- Public Knowledge Reconciliation — Parallel capability for locations, organizations, events, and contextual dates