CORDE enforces accountable disclosure in confidential AI
Original title: Enforcing Accountable Disclosure in Confidential AI
CORDE uses Rust information-flow control and TEE ledger to declare data before disclosure in confidential VMs for AI.
Advances in confidential computing allow LLMs to process prompts without exposing them. CORDE for CVMs confines user data by default but allows declared disclosure for external services or monitoring. It records exact contents, recipient, and reason in tamper-evident logs. Evaluation on NVIDIA H100 shows minimal overhead on AppWorld benchmark.
- Confine or Declare principle
- Per-user info-flow control in Rust
- TEE-based ledger with remote attestation
- 0.5% latency overhead
- Accountable disclosure for safety monitoring
From the original article
1 min read at arxiv.orgAdvances in CPU and GPU confidential computing enable confidential AI, allowing LLMs to process user prompts without exposing them to service providers or infrastructure operators. However, AI agents often need to send prompt-derived information to external services, and providers may also require limited disclosure for safety monitoring and failure diagnosis.
By Shumpei Shiina · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.