Search papers, labs, and topics across Lattice.
This paper introduces SparSEEty, a novel token extraction attack that leverages activation sparsity in large language models (LLMs) to exploit input-dependent weight accesses during inference. By constructing a neuron-activation oracle through side channels and inverting activation traces, SparSEEty successfully reconstructs input tokens with high fidelity, achieving BLEU scores above 0.95. The method addresses significant challenges in covert monitoring and accurate token reconstruction, demonstrating its effectiveness even against LLMs running in confidential virtual machines.
SparSEEty reveals that even LLMs in secure environments can be vulnerable to token extraction attacks through clever exploitation of activation sparsity.
Modern large language models (LLMs) exhibit activation sparsity, wherein only a subset of their neurons is activated for given input tokens. Researchers have leveraged this property to optimize LLM serving systems by omitting weight accesses and computations pertaining to inactive neurons. Unfortunately, however, such optimizations create input-dependent weight accesses, which can be leaked over side channels. We present SparSEEty, a new token extraction attack that exploits input-dependent neuron weight accesses introduced by sparsity-exploiting LLM serving systems. SparSEEty first constructs a neuron-activation oracle using neuron weight access side channels during LLM inference, and then inverts the activation traces to reconstruct the input tokens, forming an end-to-end token extraction attack. We instantiate SparSEEty against an LLM serving system protected inside an Intel TDX confidential virtual machine (CVM), addressing three key challenges: (i) constructing a neuron-activation oracle using a combination of side channels exposed by CVMs, (ii) reducing inference-time overheads of neuron activation monitoring for covertness, and (iii) accurately inverting partial binary activation traces back to tokens. Our evaluation shows that SparSEEty can reconstruct both prompt and response tokens with consistently high BLEU scores (>0.95) across various models and datasets, while incurring monitoring overheads of 3.7% to 7.2%.
