The Proxy Knows Too Much: Sealing LLM API Routers with Attested TEEs
Researchers have developed AEGIS, a novel API router designed to enhance the security of large language model (LLM) interactions. AEGIS utilizes attested trusted execution environments (TEEs) to ensure that the router acts as a faithful passthrough, preventing malicious actors from rewriting tool calls, injecting malicious code, or exfiltrating sensitive data. The system confines plaintext handling to a secure hardware enclave, with the client verifying the integrity of this enclave before data is processed. This approach effectively blocks known attack vectors that target plaintext-handling routers, with minimal overhead. AI
IMPACT Enhances LLM security by preventing man-in-the-middle attacks on API routers.