Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.19247 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912974969503744 |
|---|---|
| author | Shamsi, Zafir Chekuru, Nikhil Guzman, Zachary Garg, Shivank |
| author_facet | Shamsi, Zafir Chekuru, Nikhil Guzman, Zachary Garg, Shivank |
| contents | Large Language Models (LLMs) are increasingly integrated into high-stakes applications, making robust safety guarantees a central practical and commercial concern. Existing safety evaluations predominantly rely on fixed collections of harmful prompts, implicitly assuming non-adaptive adversaries and thereby overlooking realistic attack scenarios in which inputs are iteratively refined to evade safeguards. In this work, we examine the vulnerability of contemporary language models to automated, adversarial prompt refinement. We repurpose black-box prompt optimization techniques, originally designed to improve performance on benign tasks, to systematically search for safety failures. Using DSPy, we apply three such optimizers to prompts drawn from HarmfulQA and JailbreakBench, explicitly optimizing toward a continuous danger score in the range 0 to 1 provided by an independent evaluator model (GPT-5.1). Our results demonstrate a substantial reduction in effective safety safeguards, with the effects being especially pronounced for open-source small language models. For example, the average danger score of Qwen 3 8B increases from 0.09 in its baseline setting to 0.79 after optimization. These findings suggest that static benchmarks may underestimate residual risk, indicating that automated, adaptive red-teaming is a necessary component of robust safety evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_19247 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | When Prompt Optimization Becomes Jailbreaking: Adaptive Red-Teaming of Large Language Models Shamsi, Zafir Chekuru, Nikhil Guzman, Zachary Garg, Shivank Computation and Language Artificial Intelligence Large Language Models (LLMs) are increasingly integrated into high-stakes applications, making robust safety guarantees a central practical and commercial concern. Existing safety evaluations predominantly rely on fixed collections of harmful prompts, implicitly assuming non-adaptive adversaries and thereby overlooking realistic attack scenarios in which inputs are iteratively refined to evade safeguards. In this work, we examine the vulnerability of contemporary language models to automated, adversarial prompt refinement. We repurpose black-box prompt optimization techniques, originally designed to improve performance on benign tasks, to systematically search for safety failures. Using DSPy, we apply three such optimizers to prompts drawn from HarmfulQA and JailbreakBench, explicitly optimizing toward a continuous danger score in the range 0 to 1 provided by an independent evaluator model (GPT-5.1). Our results demonstrate a substantial reduction in effective safety safeguards, with the effects being especially pronounced for open-source small language models. For example, the average danger score of Qwen 3 8B increases from 0.09 in its baseline setting to 0.79 after optimization. These findings suggest that static benchmarks may underestimate residual risk, indicating that automated, adaptive red-teaming is a necessary component of robust safety evaluation. |
| title | When Prompt Optimization Becomes Jailbreaking: Adaptive Red-Teaming of Large Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2603.19247 |