Adversarial Attacks on LLM-as-a-Judge Systems: Insights from Prompt Injections
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912346544275456 |
|---|---|
| author | Maloyan, Narek Namiot, Dmitry |
| author_facet | Maloyan, Narek Namiot, Dmitry |
| contents | LLM as judge systems used to assess text quality code correctness and argument strength are vulnerable to prompt injection attacks. We introduce a framework that separates content author attacks from system prompt attacks and evaluate five models Gemma 3.27B Gemma 3.4B Llama 3.2 3B GPT 4 and Claude 3 Opus on four tasks with various defenses using fifty prompts per condition. Attacks achieved up to seventy three point eight percent success smaller models proved more vulnerable and transferability ranged from fifty point five to sixty two point six percent. Our results contrast with Universal Prompt Injection and AdvPrompter We recommend multi model committees and comparative scoring and release all code and datasets |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_18333 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Adversarial Attacks on LLM-as-a-Judge Systems: Insights from Prompt Injections Maloyan, Narek Namiot, Dmitry Cryptography and Security Computation and Language LLM as judge systems used to assess text quality code correctness and argument strength are vulnerable to prompt injection attacks. We introduce a framework that separates content author attacks from system prompt attacks and evaluate five models Gemma 3.27B Gemma 3.4B Llama 3.2 3B GPT 4 and Claude 3 Opus on four tasks with various defenses using fifty prompts per condition. Attacks achieved up to seventy three point eight percent success smaller models proved more vulnerable and transferability ranged from fifty point five to sixty two point six percent. Our results contrast with Universal Prompt Injection and AdvPrompter We recommend multi model committees and comparative scoring and release all code and datasets |
| title | Adversarial Attacks on LLM-as-a-Judge Systems: Insights from Prompt Injections |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2504.18333 |