Researchers at Weill Cornell Medicine violated privacy policies by uploading sensitive patient radiology data and clinical information into unauthorized AI tools for a research study.
Weill Cornell Medicine disclosed that protected health information from NewYork-Presbyterian facilities in New York City was processed by unidentified AI tools during radiology research without authorization. The data included medical record identifiers and clinical details. The institution suspended the research and requested deletion; it reported no evidence of unauthorized viewing or subsequent misuse.
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 incident involved uploading non-public patient clinical data and medical record numbers into unauthorized AI tools, directly compromising patient privacy.
Additional risk subdomains
- 6.5 Governance failure: The use of 'shadow AI' tools by researchers highlights a failure in institutional oversight and regulatory compliance mechanisms regarding AI deployment.
Causal factors
- Entity: Human
- Intent: Unintentional
- Timing: Post-deployment
The privacy incident was caused by human researchers uploading patient data into AI tools without authorization, which was an unintentional policy violation during the post-deployment use of these tools.
EU AI Act risk tier
High Risk: The incident involves 'radiology artificial intelligence research' and 'radiology research study', which falls under AI used in healthcare, such as diagnostic tools.
AI system and alleged parties
- AI system: unidentified AI tools
- AI purpose: Medical Diagnosis Support; Data Grouping
- Behaviour type: Tool
- Alleged developer: AI software developers
- Alleged deployer: Weill Cornell Medicine, Researchers, Research institutions, Healthcare organizations
- Alleged harmed parties: Weill Cornell Medicine radiology patients, Privacy, Patients, Biometric data subjects
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 Negligible
- Epistemic: direct Negligible, indirect Negligible
- Child sexual exploitation and abuse: direct Negligible, indirect Negligible
Privacy
Reported: The report explicitly describes a privacy incident involving patient health information used in an unauthorized capacity.
Directly caused: Patient clinical data, medical record numbers, and radiology accession numbers were uploaded to unauthorized AI tools without patient permission, violating institutional privacy policies.
Inferred additional harm: Although no misuse was detected, the exposure of protected health information to third-party AI systems represents a potential long-term privacy risk for the affected patients.
People affected
Potential causes
Management
- Insufficient Staff Training: Management did not provide adequate training on safe AI data handling.
- Lack of AI Governance Framework: No comprehensive governance structure to manage institutional AI tool adoption.
Technology
- Use of Unapproved Third-Party AI: Researchers utilized unapproved external AI tools for processing medical data.
- Lack of Technical Upload Blockers: No automated system blocks prevented uploading sensitive PHI to external AI.
- Inadequate Data Loss Prevention: DLP tools failed to detect and stop the transmission of patient health data.
Data Inputs
- Input of Protected Health Info: Sensitive patient clinical data and MRNs were fed directly into the AI tool.
- Lack of Data De-identification: Data was not properly anonymized before being entered into the AI system.
Human Factors
- Unauthorized Researcher Actions: Staff uploaded sensitive patient data without obtaining proper authorization.
- Lack of Policy Awareness: Researchers were unaware of or ignored WCM policies regarding AI tools.
Process and Methods
- Inadequate Compliance Oversight: Lack of active monitoring to detect policy violations during AI research.
- No Pre-use AI Tool Review: No formal process was enforced to review AI tools prior to research use.
Regulatory Environment
- Strict Privacy Regulations: HIPAA and institutional policies prohibit unauthorized sharing of PHI.
- Rapidly Evolving AI Regulations: Lag in clear regulatory frameworks specifically targeting shadow AI in health.
Information quality
- Classification confidence: High
- Reason for confidence: The reports from multiple sources consistently describe the timeline, the specific types of patient data involved, and the institutional response. While the exact number of affected patients and the specific AI tools are not disclosed, the core facts of the privacy policy violation are clear.
- Ambiguities identified: The specific AI tools used and the exact number of affected patients are not disclosed.
- Alternative interpretations: None.
Researchers at Weill Cornell Medicine violated privacy policies by uploading patient radiology and clinical data into unauthorized third-party AI tools. The incident represents a localized data governance and privacy failure within an academic institution, carrying negligible implications for national security.
- Overall national security impact: Negligible
- Scope: Single nation
- Primary target: United States
- Alleged perpetrator: Weill Cornell Medicine researchers
Threat characteristics
- Imminence: Long-term. The incident is a past event that has been contained through research suspension and data deletion requests, presenting no immediate threat.
- Autonomy: Human-controlled. The AI tools were used as passive instruments to process data manually uploaded by human researchers.
- Novelty: Established threat. The unauthorized use of third-party cloud and AI services (shadow AI) to process sensitive data is a well-known and common institutional risk.
Impact by dimension
- Physical security: Negligible. The incident involved researchers uploading patient data to AI tools and did not affect physical systems, critical infrastructure, or human safety.
- Information security: Negligible. No classified information, intelligence capabilities, or state-level information operations were involved or compromised.
- Sovereignty: Negligible. The incident was a localized academic policy violation with no impact on state authority, elections, or government decision-making.
- Economic security: Negligible. There was no theft of strategic technology, financial system manipulation, or loss of national AI competitive advantage.
- Societal stability: Negligible. While patient privacy was compromised, the incident was localized to an academic study with no population-scale societal instability or systemic civil rights violations.