Operationalizing CaMeL: Strengthening LLM Defenses for Enterprise Deployment
Journal:
arXiv
Published Date:
May 28, 2025
Abstract
CaMeL (Capabilities for Machine Learning) introduces a capability-based
sandbox to mitigate prompt injection attacks in large language model (LLM)
agents. While effective, CaMeL assumes a trusted user prompt, omits
side-channel concerns, and incurs performance tradeoffs due to its dual-LLM
design. This response identifies these issues and proposes engineering
improvements to expand CaMeL's threat coverage and operational usability. We
introduce: (1) prompt screening for initial inputs, (2) output auditing to
detect instruction leakage, (3) a tiered-risk access model to balance usability
and control, and (4) a verified intermediate language for formal guarantees.
Together, these upgrades align CaMeL with best practices in enterprise security
and support scalable deployment.