Facebook's 'People You May Know' feature inappropriately recommended patients to their psychiatrist and to each other, potentially exposing sensitive mental health information and violating patient privacy.
Facebook's 'People You May Know' recommendation algorithm began suggesting patients as friends to their psychiatrist Lisa, and recommending fellow patients to each other. Lisa, a psychiatrist who infrequently used Facebook, noticed that her patients - mostly senior citizens and people with serious health or developmental issues - were appearing in her friend recommendations despite not sharing contacts or having any obvious digital connections. One patient, a 30-something snowboarder, began seeing recommendations for elderly and disabled individuals he recognized as fellow patients from the psychiatrist's office. Another female patient received a friend recommendation for a fellow patient she had only seen in the office elevator, suddenly learning their full name and profile information. Lisa had not friended any patients or looked up their profiles, and there was no shared guest wifi network. The most likely explanation was that patients had Lisa's phone number in their contacts and had given Facebook access to their phone contacts, allowing the algorithm to infer connections between all parties. Lisa expressed serious concern about this 'massive privacy fail' given that her patients include people with HIV, suicide attempts, and women in violent relationships. The medical community began recommending patients not log into social media at medical offices or even leave phones in cars during appointments.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that memorize and leak sensitive personal data or infer private information about individuals without their consent. Unexpected or unauthorized sharing of data and information can compromise user expectation of privacy, assist identity theft, or cause loss of confidential intellectual property.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed