Facebook's Friend Suggestion Feature Recommends Patients of Psychiatrist to Each Other

Facebook's "People You May Know" (PYMK) feature was reported by a psychiatrist for recommending her patients as friends through recommendations, violating patients' privacy and confidentiality.

Risk classification

• Primary risk domain: 2 Privacy & Security
• Primary risk subdomain: 2.1 Compromise of privacy by obtaining, leaking or correctly inferring sensitive information

The AI system inferred sensitive, non-public relationships between psychiatric patients and leaked their identities to one another based on shared contact data.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The privacy breach was caused by the automated decisions of Facebook's recommendation algorithm operating in production, which produced unintended and harmful social connections.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The system is a social media recommendation feature, which generally falls under minimal risk applications as it does not fit the high-risk or unacceptable categories.

AI system and alleged parties

• AI system: People You May Know (Meta)
• AI purpose: Matchmaking; Content Recommendation
• Behaviour type: Autonomous
• Alleged developer: Facebook
• Alleged deployer: Facebook
• Alleged harmed parties: pseudonymized psychiatrist's patients, pseudonymized psychiatrist, patients, healthcare providers

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Minor, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Privacy

Reported: Yes, the report explicitly describes a 'massive privacy fail' where patients' identities and medical associations were exposed.

Directly caused: The algorithm exposed personally identifiable information (full names and profiles) of patients to other patients of the same psychiatrist.

Inferred additional harm: Many other patients of this psychiatrist, and potentially patients of other medical professionals, likely had their private associations exposed due to contact-syncing.

Psychological

Reported: Yes, the report describes the psychiatrist feeling awkward and silent, and notes the severe distress potential for vulnerable patients.

Directly caused: The psychiatrist experienced professional distress and awkwardness when confronted with the algorithm's recommendations.

Indirectly caused: Patients with sensitive conditions experienced distress and anxiety due to the exposure of their private medical associations.

Inferred additional harm: It is highly likely that other patients in the practice experienced anxiety and fear of exposure upon seeing fellow patients recommended to them.

People affected

• Occurrences reported: 1
• People reportedly harmed: 6
• People reportedly exposed: 10

Potential causes

Management

• Prioritizing Growth Over Privacy: Maximizing network growth and engagement over user data privacy.

Technology

• Context-Insensitive Algorithm: Algorithm treats medical contacts like ordinary social connections.
• Implicit Network Mapping: System automatically links users sharing a common contact.

Data Inputs

• Imported Contact Lists: Users uploading phone contacts containing the psychiatrist's number.
• Shared Phone Number Data: Psychiatrist providing cell phone number to Facebook.

Human Factors

• Unawareness of Contact Sharing Risks: Users did not realize contact sharing would link them to other patients.
• Overreliance on Default Privacy: Assuming private medical relationships would remain confidential.

Process and Methods

• Inadequate Privacy Risk Assessment: Lack of evaluation on how contact matching affects sensitive professions.
• Opaque Recommendation Logic: Lack of transparency in how friend recommendations are generated.

Regulatory Environment

• Absence of Health Privacy Mandates: Lack of regulations preventing social networks from mapping patient networks.

Information quality

• Classification confidence: High
• Reason for confidence: The report provides a clear, detailed account of the privacy failure, including the mechanism (contact syncing of the psychiatrist's phone number) and the specific real-world consequences for patients.
• Ambiguities identified: Facebook did not officially confirm the exact technical cause of the recommendations due to privacy restrictions.
• Alternative interpretations: None that are highly plausible; the contact-syncing explanation is strongly supported by the evidence.

This incident involves a localized privacy breach where Facebook's autonomous recommendation algorithm exposed patient identities to each other via shared phone contacts. While it highlights critical algorithmic privacy risks for sensitive consumer data, it has negligible national security implications across all core assessment categories.

• Overall national security impact: Negligible
• Scope: Single nation
• Primary target: No clear primary
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The incident represents an ongoing data privacy and algorithmic design concern rather than an active, imminent national security crisis.
• Autonomy: Full autonomy. The 'People You May Know' recommendation algorithm operates fully autonomously, generating and displaying friend suggestions without human intervention.
• Novelty: Evolved capability. Represents an evolved capability where standard social graph mapping algorithms inadvertently expose sensitive real-world relationships and medical associations.

Impact by dimension

• Physical security: Negligible. No physical security, critical infrastructure, or kinetic threats are present in this consumer social media privacy incident.
• Information security: Negligible. The incident involves consumer data privacy leaks on a commercial platform, with no connection to intelligence capabilities, state-sponsored information warfare, or classified systems.
• Sovereignty: Negligible. No sovereign territory, electoral processes, or government decision-making systems were impacted by this commercial algorithm behavior.
• Economic security: Negligible. The incident does not involve strategic technology theft, financial infrastructure attacks, or threats to national economic security.
• Societal stability: Negligible. While the privacy breach caused distress to vulnerable patients, it was localized to a single practice and does not threaten population-scale societal stability or civil liberties.