Reproduce “Enforcing Accountable Disclosure in Confidential AI” 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

Shumpei Shiina, Haruka Takahashi, Kento Sasaki, Takuya Hayashi, Tsubasa Takahashi

CORDE records user-authorized disclosures before external calls, combining Rust-based information-flow controls with a TEE-protected ledger bound to the inference service. Each record specifies the payload, recipient, and reason. In a synthetic prompt-injection demonstration, the agent still sends a confidential reference, but the user can inspect the disclosure. Across 723 sequential AppWorld tasks, mean latency was 154.34 seconds versus 153.66 seconds for the baseline.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.