Search papers, labs, and topics across Lattice.
The paper introduces Back-Reveal, a novel data exfiltration attack targeting tool-use LLM agents by embedding semantic triggers into fine-tuned models. This attack leverages memory-access tool calls to retrieve and exfiltrate user context through disguised retrieval tool calls. Experiments demonstrate that multi-turn interactions amplify the impact, enabling sustained information leakage via attacker-controlled retrieval responses.
Backdoored LLM agents can stealthily leak your sensitive data via disguised tool calls, and the risk grows with each turn of interaction.
Tool-use large language model (LLM) agents are increasingly deployed to support sensitive workflows, relying on tool calls for retrieval, external API access, and session memory management. While prior research has examined various threats, the risk of systematic data exfiltration by backdoored agents remains underexplored. In this work, we present Back-Reveal, a data exfiltration attack that embeds semantic triggers into fine-tuned LLM agents. When triggered, the backdoored agent invokes memory-access tool calls to retrieve stored user context and exfiltrates it via disguised retrieval tool calls. We further demonstrate that multi-turn interaction amplifies the impact of data exfiltration, as attacker-controlled retrieval responses can subtly steer subsequent agent behavior and user interactions, enabling sustained and cumulative information leakage over time. Our experimental results expose a critical vulnerability in LLM agents with tool access and highlight the need for defenses against exfiltration-oriented backdoors.
