Reproduce “$π$Creds: Privately Inferred Credentials” 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
Samuel Breckenridge, Dani Vilardell, Derek Leung, Andrés Fábrega, James Austgen, Farinaz Koushanfar, Ari Juels
A system called Privately Inferred Credentials (𝜋Creds) combines Large Language Models and Trusted Execution Environments to enable verifiable attestations derived from unstructured data and concerning proprietary software. The framework supports richer semantic claims and characterizes new application-level security threats, such as Source-Constrained Adversarial Examples and Authenticated Covert Predicate Poisoning, inherent to LLM integration.
Key insights
- LLMs significantly expand the expressiveness of verifiable claims by enabling semantic reasoning over diverse unstructured data sources for generating credentials.
- Combining LLMs and TEEs allows for attesting to properties of proprietary software without disclosing its source code, creating a new domain for verifiable credentials.
- The integration of LLMs introduces novel application-level security threats, specifically Source-Constrained Adversarial Examples (SCAE) and Authenticated Covert Predicate Poisoning (ACPP), which require formalization and empirical analysis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.