Privacy & Security
AI systems can pose risks to privacy through data collection and processing, and may contain vulnerabilities that can be exploited by attackers.
Risk Breakdown
Monthly Incidents
Compromise of privacy
Represents 55% of Privacy & Security risks
Examples:
- Unauthorized data collection
- Re-identification of anonymized data
- Inference attacks revealing sensitive information
AI system vulnerabilities and attacks
Represents 45% of Privacy & Security risks
Examples:
- Adversarial attacks on machine learning models
- Data poisoning during training
- Model extraction and theft
Related Incidents
Healthcare Data Leak
Date: 2023-06-08Impact: CriticalStatus: Resolved
An AI system used for patient diagnosis was found to be leaking sensitive medical information that could be used to identify individuals.
Model Extraction Attack
Date: 2023-05-17Impact: HighStatus: Mitigated
Researchers demonstrated how a proprietary AI model could be stolen through carefully crafted queries that revealed its internal parameters.
Adversarial Attack on Authentication
Date: 2023-04-22Impact: CriticalStatus: Resolved
A facial recognition authentication system was bypassed using adversarial examples generated by another AI system.
Data Poisoning Incident
Date: 2023-03-11Impact: MediumStatus: Under Investigation
A public-facing AI model was compromised when malicious actors contributed tainted data to its continuous learning pipeline.
Mitigation Strategies
- Privacy-preserving machine learning techniques
- Differential privacy implementation
- Robust security testing
- Adversarial training
- Regular security audits