Jailbreaking is (Mostly) Simpler Than You Think
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916646335021056 |
|---|---|
| author | Russinovich, Mark Salem, Ahmed |
| author_facet | Russinovich, Mark Salem, Ahmed |
| contents | We introduce the Context Compliance Attack (CCA), a novel, optimization-free method for bypassing AI safety mechanisms. Unlike current approaches -- which rely on complex prompt engineering and computationally intensive optimization -- CCA exploits a fundamental architectural vulnerability inherent in many deployed AI systems. By subtly manipulating conversation history, CCA convinces the model to comply with a fabricated dialogue context, thereby triggering restricted behavior. Our evaluation across a diverse set of open-source and proprietary models demonstrates that this simple attack can circumvent state-of-the-art safety protocols. We discuss the implications of these findings and propose practical mitigation strategies to fortify AI systems against such elementary yet effective adversarial tactics. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_05264 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Jailbreaking is (Mostly) Simpler Than You Think Russinovich, Mark Salem, Ahmed Cryptography and Security Artificial Intelligence We introduce the Context Compliance Attack (CCA), a novel, optimization-free method for bypassing AI safety mechanisms. Unlike current approaches -- which rely on complex prompt engineering and computationally intensive optimization -- CCA exploits a fundamental architectural vulnerability inherent in many deployed AI systems. By subtly manipulating conversation history, CCA convinces the model to comply with a fabricated dialogue context, thereby triggering restricted behavior. Our evaluation across a diverse set of open-source and proprietary models demonstrates that this simple attack can circumvent state-of-the-art safety protocols. We discuss the implications of these findings and propose practical mitigation strategies to fortify AI systems against such elementary yet effective adversarial tactics. |
| title | Jailbreaking is (Mostly) Simpler Than You Think |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2503.05264 |