Data No One Gave: Inference and Reconstruction in Generative AI
Duration: 60 Minutes
24 Days Left To REGISTER
Id: 213372
Eran Kahana
Overview
This session examines the privacy and legal challenges surrounding data that AI systems infer or reconstruct rather than data users explicitly provide. It explains how models draw statistical inferences about individuals from non-personal or aggregated inputs and explores the risks of model inversion, membership inference, and synthetic-data generation.
The program analyzes the regulatory treatment of inferred data-the GDPR definition of personal data as applied to inferred identities, and CCPA rights to know, delete, and correct inferred information-and the ethical problems of consent, discriminatory profiling, and the utility-privacy trade-off. It closes with compliance strategies built on differential privacy, privacy-enhancing technologies, and inference-focused impact assessments.
Why you should Attend
If your models generate sensitive conclusions about individuals that were never collected, you may be processing regulated personal data without knowing it-and without a consent basis for it. Regulators are extending General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA) concepts to inferred identities, and synthetic data offers only partial cover. This session clarifies when inference becomes a legal exposure and what privacy-enhancing controls actually reduce it.
Areas Covered in the Session
- How machine-learning models infer individual attributes from non-personal or aggregated data
- Model inversion and membership inference attacks
- Synthetic data generation and its privacy implications
- The GDPR definition of personal data applied to inferred identities
- CCPA rights to know, delete, and correct inferred information
- The ambiguity of consent when data is inferred rather than collected
- Discriminatory inferences and algorithmic profiling
- Balancing utility and privacy in synthetic datasets
- Compliance strategies: differential privacy, privacy-enhancing technologies (PETs), and inference-focused Data Protection Impact Assessments (DPIAs)
Who Will Benefit
- Chief Privacy Officers and Data Protection Officers
- General Counsel and In-House Counsel
- Chief Data and AI Officers
- Privacy Engineers
- Compliance and Risk Officers
- AI and Machine-Learning Researchers and Leads
- Product Counsel
Speaker Profile
In his practice, Eran counsels clients on a wide variety of matters related to AI, cybersecurity, privacy, technology law, trademarks, patents, and copyright issues. Eran also serves in a variety of cybersecurity thought leadership roles and works closely with the FBI, Department of Justice, Secret Service, and colleagues from the private and academic sectors to set, promote, and sustain cybersecurity best practices.
At Stanford Law School, Eran writes and lectures on the intersect between law and AI and is a frequent speaker at Stanford's annual Digital Economy Best Practices Conference. He has been cited in Oxford University Professor Marcus Du Satoy’s book The Creativity Code: Art and Innovation in the Age of AI and has been interviewed on AI, cybersecurity, privacy, and technology law by Bloomberg Law, BBC, Canadian Broadcasting Corporation (CBC) radio, KABC radio, Minnesota Public Radio, Twin Cities Business magazine, Star Tribune, Minnesota Lawyer, TheStreet.com, Quartz magazine, KARE 11, and Stanford University Radio, KZSU FM.
Access Recorded Version
One Attendee / Group Attendees
Unlimited Viewing Recorded Version for 6 months ( Access information will be emailed 24 hours after the completion of live webinar)