What You Approve Is What Executes: Consent Integrity for Black-Box LLM Agents
For developers and users of LLM-based coding agents, this work formalizes a critical security property and provides an honest assessment of the inherent trade-offs, though the defense is not fully implemented.
The paper identifies a vulnerability in LLM coding agents where a compromised agent can forge approval summaries (Lies-in-the-Loop attack), and proposes Consent Integrity, a property requiring a trusted mediator to render actions for human approval. A prototype on GTFOBins silently passes 10.0% of commands and marks 87.0% as uninspectable on normal commands, highlighting a trade-off between trust and over-prompting.
Coding agents gate consequential actions behind a human-in-the-loop approval dialog, but the dialog is narrated by the agent itself: the human approves a summary the agent writes. The Lies-in-the-Loop (LITL) attack shows that summary is forgeable, so a compromised agent can show a benign description while a different action runs. This paper names the missing property, Consent Integrity, by importing What You See Is What You Sign (WYSIWYS) and the trusted-path property into the agent approval channel: the action shown to the human must be rendered by a trusted mediator from the real action at the boundary, not the agent's narration, over a path the agent cannot spoof, and bound to the exact action that executes. Two twists distinguish it from classical WYSIWYS: the renderer is the adversary, and the boundary ground truth is a low-level event that must be decoded without trusting the agent. Since no decoder is complete, the realizable target is analyzer-relative: whatever the analyzer cannot classify is surfaced as uninspectable rather than silently approved. A prototype implements the analyzer, renderer, and bind-to-execution; total mediation and the trusted path are specified but assumed, not implemented. On GTFOBins, an independent corpus of 1330 trusted-tool abuses, the prototype silently passes 10.0% (every instance through a trusted tool); on tldr, 28,798 normal-usage commands, it marks 87.0% uninspectable. These two independent measurements bracket the design's central tension: the trust list that bounds silent passes is the same one that drives over-prompting, and a boundary-only mediator can move along that frontier but not escape it. The contribution is the property, the mechanism, and an honest position on that frontier, not a solved defense.