Taxonomy identifies nine dimensions of prompt injection attacks
Original title: The Devil is in the Context: Nine Dimensions of Prompt Injection Attacks
The taxonomy covers six real-world vulnerabilities and reveals gaps in nine existing benchmarks.
Prompt injection varies along nine dimensions that affect injected instructions, relation to user task, and system manifestation. The taxonomy characterizes six real-world vulnerabilities and nine benchmarks, showing most cover only a small region of the space. It supports automated annotation and targeted test case generation.
- Nine dimensions describe prompt injection variations
- Six real-world vulnerabilities characterized
- 13% of benchmark category pairs remain untested
From the original article
1 min read at arxiv.orgPrompt injection is usually discussed and evaluated as a single, homogeneous type of attack. However, vulnerabilities reported in real-world systems show that prompt injections vary widely. Moreover, the same instruction can be a prompt injection in one context and a legitimate request in another, depending on the user's intent and the system's policies.
By Jiamin Chang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.