Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore an open-source framework that enables private data analytics for public research using Trusted Execution Environments (TEEs) in this 30-minute conference talk. Learn how ManaTEE supports verifiable AI transparency, particularly for proprietary models where open inspection isn't feasible, by allowing sensitive model evaluation and data analysis in isolated, secure environments with cryptographic attestation that ensures the integrity of both models and data. Discover how TEEs enable external researchers and auditors to assess AI systems without requiring direct access to the model or sensitive data, while ManaTEE simplifies this process through a secure, interactive Jupyter Notebook interface where users can load benchmarks, write evaluation code, and analyze results while preserving data confidentiality and model secrecy. Watch a demonstration of how ManaTEE evaluates a closed-source AI model in a reproducible and auditable way, helping to balance the critical need for transparency with confidentiality requirements in AI systems.
