Evidence: well sourced. Imported from the supplied 65-Case Master Edition, dated September 19, 2026. Source links and classifications are retained as an attributed case account; import is not an independent source review.

Case at a glance

Case number
024
Date / range
2008
Sector
Re-identification and inference
Genetic asset
A public genome with a redacted APOE region
Security principle
Correlated-Data Leakage

Event summary

James Watson withheld the APOE region from his public genome because of its association with Alzheimer disease risk, but researchers showed that nearby correlated variants could still support inference. Deleting the secret field may not delete the secret.

Source: pubmed.ncbi.nlm.nih.gov — James Watson's Redacted APOE Region source 1.

Source: pmc.ncbi.nlm.nih.gov — James Watson's Redacted APOE Region source 2.

The case in context

The Watson example concerns selective disclosure. Withholding one region of a genome did not necessarily remove the information needed to infer something about that region, because visible variants could remain correlated with the hidden data. The researchers did not disclose his status.

Redaction therefore needs to consider relationships between fields, not merely whether the sensitive field is absent. This is a privacy lesson rather than a diagnosis: inferred genetic risk does not establish disease or destiny. A release should be assessed for what its remaining information supports, including in combination with outside reference data.

Acquisition and processing

public genome with narrow redaction → correlated flanking variants → statistical inference → sensitive genotype risk

The sequence of events

  1. public genome with narrow redaction
  2. correlated flanking variants
  3. statistical inference
  4. sensitive genotype risk

What became inferable or exposed

A public genome with a redacted APOE region

James Watson withheld the APOE region from his public genome because of its association with Alzheimer disease risk, but researchers showed that nearby correlated variants could still support inference. Deleting the secret field may not delete the secret.

Security dimensions

Confidentiality

The confidentiality question concerns a public genome with a redacted apoe region. Exposure and further inference must be distinguished from the fact of collection or availability.

Integrity

The integrity question is whether the described material, permissions, processing, or interpretation can be relied upon. Correlated-Data Leakage identifies the particular boundary examined here.

Availability

Access and continuity are assessed for the described event; potential effects are not presented as confirmed outages or losses.

Provenance

The relevant chain follows a public genome with a redacted apoe region through the stages shown below. Missing public detail is not proof that internal records did not exist.

GeneticSecurity.org analysis

Genetic Exposure Radius

Not assessed

No single level is assigned where the supplied dossier gives a range, conditional outcome, or broad institutional consequence. The affected parties and proposed assessment are shown separately.

Confidence: not assigned. Classification: GeneticSecurity.org analysis.

Genetic Persistence Risk

Not assessed

Persistence depends on the specific biological material or information retained. A potential effect is not treated as an observed genomic disclosure.

Confidence: not assigned. Classification: GeneticSecurity.org analysis.

Genetic Provenance Integrity

Not assessed

A numeric provenance level is not inferred from the existence of a source or court record. It requires evidence of the relevant custody and processing controls.

Confidence: not assigned. Classification: GeneticSecurity.org analysis.

Proposed classification and its limits

Suggested GER: GER-1/2 because APOE information may matter to relatives. Suggested GPR: GPR-5. GPI: not central.

These are proposed classifications from the supplied case dossier. Conditional scores describe an assumed exposure; they are not evidence that it occurred. A single numeric value is left unassigned when the asset or outcome is not sufficiently bounded.

What this case does not prove

The example does not establish an Alzheimer diagnosis, a predetermined outcome, or perfect genotype inference.

Mitigations and lessons

  • Correlation-aware redaction
  • Privacy threat modeling before release
  • Family-impact counseling
  • Controlled rather than public access
  • Periodic reassessment as reference panels improve

Primary sources

Secondary sources

No additional source listed. See the evidence notes for limitations.

Policy and standards

Genetic Security Policy and Standards

Review and correction history

Source edition: September 19, 2026. Imported case account; no substantive corrections recorded.

Correction policy and log

Cite this case

GS-CASE-024. The Gene He Hid: Why Genomic Redaction Is Hard. GeneticSecurity.org. https://geneticsecurity.org/cases/024-watson-apoe-genome-redaction-failure/