CyLab Seminar: Taeho Jung

Speaker: Taeho Jung Dennis O. Doughty Collegiate Professor of Data Security and Privacy Notre Dame University Talk Title: Security, Privacy, and Trust in the AI Era: Towards a secure zero-trust AI ecosystem Abstract: As AI systems become better at using our data — prompts, screens, records, and entire workflows — a central question emerges: who keeps control of the data? The most valuable data is also the most regulated, scattered across thousands of hospitals, banks, and enterprises that cannot legally or practically pool it in plaintext. This talk presents a research program aimed at a structure in which using data does not mean surrendering it. I will present results along three complementary threads. First, private set computation over encrypted data: fully homomorphic encryption protocols that perform exact private set intersection over a billion records distributed across a thousand encrypted silos in about a minute, fuzzy (cosine/Hamming) matching over high-dimensional embeddings up to three orders of magnitude faster than prior work, and the first encrypted label selection-and-analytics protocol enabling end-to-end encrypted fraud and AML analytics across thousands of institutions. Second, minimal trusted hardware: we show that naively placing a differential-privacy sampler in an SGX enclave lets the host recover over 99% of the noise draws, and then design systems in which small, verified enclaves stay off the hot path — scaling secure federated-learning aggregation to 50,000 clients and making Intel TDX confidential VMs continuously and anonymously attestable at cloud scale. Third, Web3DB: a decentralized database series in which only an enclave is trusted, providing owner-defined row-level access control, near-cloud analytics performance, and serializable transactions over blockchain and IPFS. I will close with a vision that composes these primitives — policy, computation, and evidence layers — into a secure zero-trust ecosystem for agentic AI. Bio: Taeho