Whisper Leak

Capture Summary

Paper introducing a side-channel attack that infers user prompt topics from encrypted LLM traffic by analyzing packet size and timing patterns in streaming responses. Search result reports testing across 28 popular LLMs and mitigations such as random padding, token batching, and packet injection.

Relevance

• Expands Security for AI beyond prompt injection into deployment privacy and network metadata leakage.
• Useful for research on privacy controls for streaming LLM APIs.
• Connects AI system security with traditional side-channel threat modeling.

Collection Notes

Collected as current privacy/security paper.