Reproduce “Private, Verifiable, and Auditable AI Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tobin South
Tobin South's doctoral dissertation presents an integrated framework for engineering trustworthy AI systems, particularly foundation models and LLMs, by addressing privacy, verifiability, and auditability across their lifecycle. The work combines cryptographic proofs, confidential computing, and identity management to provide concrete solutions for secure and accountable AI deployment.
Key insights
- Achieving comprehensive security, verifiability, and auditability in AI requires a cohesive integration of cryptographic proofs, confidential computing, and identity frameworks, rather than isolated solutions.
- Retrieval Augmented Generation (RAG) fundamentally enhances LLM auditability and updatability by grounding responses in specific, traceable external data sources, reducing reliance on opaque model memorization.
- Practical deployment of verifiable AI systems can be achieved by decoupling expensive proof generation from real-time inference (a "predict, then prove" strategy) and proving only critical sub-components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.