Facebook's 'People You May Know' (PYMK) recommendation algorithm inadvertently outed sex workers by suggesting their private accounts to clients or family members. The system used opaque data signals to link separate identities, causing significant privacy and safety concerns for users who intentionally kept their professional and personal lives separate. Facebook failed to provide clear explanations for these linkages or a viable opt-out mechanism, and even provided incorrect information regarding potential workarounds.
Facebook's "People You May Know" feature reportedly outed sex workers by recommending clients to their personal accounts or family members to their business accounts with no option to opt out.
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 PYMK algorithm correctly inferred and linked highly sensitive, non-public connections between users' real-world identities and their separate sex-work personas without their consent.
Additional risk subdomains
- 7.4 Lack of transparency or interpretability: Facebook maintains an algorithmic black box around PYMK, refusing to disclose the 100-plus signals used, leaving users unable to understand or prevent the privacy leaks.
Causal factors
- Entity: AI
- Intent: Unintentional
- Timing: Post-deployment
The privacy violations were caused by the autonomous operations of Facebook's PYMK recommendation algorithm, which produced unexpected and harmful identity linkages after deployment.
EU AI Act risk tier
- Risk tier: 4 Minimal or No Risk
Minimal or No Risk: The system is a social media recommendation algorithm used for suggesting friend connections, which does not fall under the high-risk or limited-risk categories of the EU AI Act.
AI system and alleged parties
- AI system: People You May Know (Meta)
- AI purpose: Content Recommendation; Identification
- Behaviour type: Autonomous
- Alleged developer: Facebook
- Alleged deployer: Facebook
- Alleged harmed parties: sex workers using Facebook
Harm severity
Highest direct severity in any category: Substantial. 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 Minor, indirect Minor
- Democracy: direct Negligible, indirect Negligible
- Privacy: direct Minor, indirect Negligible
- Psychological: direct Minor, indirect Negligible
- Epistemic: direct Negligible, indirect Negligible
- Child sexual exploitation and abuse: direct Negligible, indirect Negligible
Civil rights
Reported: The report explicitly describes violations of human and civil rights, specifically the right to privacy and personal safety.
Directly caused: The algorithm violated the basic privacy rights of [PERSON_001] and [PERSON_004] by linking and exposing their separate identities without consent.
Indirectly caused: [PERSON_004] was harassed by fans who used the leaked information to track down her family, violating her right to safety.
Inferred additional harm: Systematic violation of privacy rights for thousands of users who rely on pseudonymity for their safety, including activists and domestic violence survivors.
Privacy
Reported: The report explicitly describes severe privacy violations caused directly by the incident.
Directly caused: Facebook's PYMK algorithm linked the private, separate identities of [PERSON_001] and [PERSON_004] using opaque data signals, exposing their real names and accounts to clients and fans without consent.
Indirectly caused: N/A
Inferred additional harm: Thousands of other sex workers and vulnerable individuals likely had their private identities linked and suggested to people in their personal or professional lives, violating their expectation of privacy.
Psychological
Reported: The report explicitly describes severe psychological distress, anxiety, and fear of exposure caused directly by the incident.
Directly caused: [PERSON_001] experienced shock and horror at seeing her clients recommended, describing the potential exposure of her real identity as a 'worst nightmare'. [PERSON_004] experienced harassment from fans who tracked down her family, causing significant distress.
Indirectly caused: N/A
Inferred additional harm: Widespread anxiety and paranoia likely affected thousands of other sex workers and vulnerable individuals who realized their private identities were being algorithmically linked and potentially exposed.
People affected
- Occurrences reported: 1
- People reportedly harmed: 5
- People reportedly exposed: 100
Potential causes
Management
- Growth-Above-All Strategy: Management prioritized network growth and density over user safety and privacy.
- Lack of Transparency: Management refused to disclose the specific signals used for recommendations.
Technology
- Black Box Recommendation Engine: PYMK uses over 100 opaque signals to link users without their consent.
- Intrusive Network Analysis: Algorithms aggressively map social connections, bypassing user privacy shields.
Data Inputs
- Undisclosed Contact Tracking: System utilizes indirect contact lists and metadata to establish linkages.
- Lack of Data Separation: Platform aggregates disparate data points, blending separate user personas.
Human Factors
- False Sense of Security: Users assumed separate emails and phone numbers would guarantee anonymity.
- User Vulnerability to Stalking: Sex workers and performers face heightened physical and professional risks.
Process and Methods
- No Opt-Out Mechanism: Facebook did not provide an option to disable or opt out of PYMK.
- Inadequate Feedback Loops: X-ing out suggestions does not prevent the user from being suggested.
- Flawed Privacy Workarounds: Documented workarounds to disable suggestions were non-existent or false.
Regulatory Environment
- Lack of Privacy Regulations: Absence of strict laws governing algorithmic contact linking and profiling.
Information quality
- Classification confidence: High
- Reason for confidence: The report provides detailed, first-hand accounts from affected individuals ([PERSON_001] and [PERSON_004]) and representatives of advocacy groups, clearly establishing the connection between Facebook's PYMK algorithm and the privacy violations. While the exact technical signals used by the algorithm remain undisclosed by Facebook, the real-world impact and the failure of the platform to provide workarounds are well-documented.
- Ambiguities identified: The exact data signals and tracking mechanisms (e.g., location services, contact scraping, or shadow profiles) used by Facebook to link the separate identities remain unconfirmed due to company secrecy.
- Alternative interpretations: None. The link between the algorithm's recommendations and the exposure of separate identities is clear and undisputed, even if the underlying technical mechanism is opaque.
Facebook's PYMK recommendation algorithm unintentionally linked and exposed the separate identities of users who sought to keep their personal and professional lives apart, such as sex workers. While posing significant privacy and safety concerns for vulnerable individuals, the incident does not present direct or substantial threats to national security, representing a minor impact on societal stability and privacy rights.
- Overall national security impact: Minor
- Response level: Moderate
- Scope: Single nation
- Primary target: No clear primary
- Alleged perpetrator: Unknown
Threat characteristics
- Imminence: Long-term. Reflects an ongoing, systemic privacy vulnerability in commercial algorithms rather than an active, fast-moving crisis.
- Autonomy: Full autonomy. The recommendation engine processed data and generated friend suggestions automatically without direct human intervention.
- Novelty: Evolved capability. Represents an evolved capability where opaque social graph algorithms link distinct user identities using complex, indirect signals.
Impact by dimension
- Physical security: Negligible. No kinetic threats, physical attacks, or critical infrastructure disruptions were caused by this commercial recommendation algorithm.
- Information security: Negligible. No state-sponsored information warfare, intelligence compromise, or classified data theft occurred as a result of this incident.
- Sovereignty: Negligible. The incident did not impact state authority, electoral systems, border control, or core government decision-making processes.
- Economic security: Negligible. No strategic technology theft, financial system attacks, or critical supply chain disruptions occurred.
- Societal stability: Minor. The incident represents a notable erosion of privacy for vulnerable populations relying on pseudonymity, but lacks broader national security threats to societal cohesion.