Prompt Injections Against Closed-Weights Models

Title: Computing Optimization-Based Prompt Injections Against Closed-Weights Models By Misusing a Fine-Tuning API

Abstract: LLMs are evolving into agentic systems, which operate a combination of trusted and untrusted input on our behalf. It is important to understand and mitigate the risk of prompt injection attacks, steering the model output towards an attacker-controlled action with a specially crafted perturbation of the input. For that, we want to understand if strong automated prompt injection attacks are possible. We surface a new threat to closed-weight Large Language Models (LLMs) that enables an attacker to compute optimization-based prompt injections. Specifically, we characterize how an attacker can leverage the loss-like information returned from the remote fine-tuning interface to guide the search for adversarial prompts. The fine-tuning interface is hosted by an LLM vendor and allows developers to fine-tune LLMs for their tasks, thus providing utility, but also exposes enough information for an attacker to compute adversarial prompts. Through an experimental analysis, we characterize the loss-like values returned by the Gemini fine-tuning API and demonstrate that they provide a useful signal for discrete optimization of adversarial prompts using a greedy search algorithm. Using the PurpleLlama prompt injection benchmark, we demonstrate attack success rates between 65% and 82% on Google's Gemini family of LLMs. These attacks exploit the classic utility-security tradeoff - the fine-tuning interface provides a useful feature for developers but also exposes the LLMs to powerful attacks.

Presenter: Andrey is a security researcher, a whitehat hacker, and a former Facebook Security engineer: website

Paper: link
Twitter: link