InfoResearchPeer-reviewedLLM-specific
The Evolution of Textual Jailbreak Attacks and Defenses for Large Language Models
- Published
- Record updated
Summary
This review examines textual jailbreak attacks and defenses for large language models under a unified task definition. It finds that historical semantic mappings and earlier assistant responses can change the meaning of a current request, so inspecting only the current input or output may miss concealed harmful intent. It also compares black-box and white-box defenses, noting that white-box methods depend on access permissions and cross-model calibration.
Topics
Related items
- InfoDressed in Authority: Practical Black-Box Jailbreaking of MLLMs via Visual Trust BiasSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
- InfoEGS: A Complementary Safety Framework for Educational LLMsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
- InfoCat and Mouse in the LLM Era: An Investigation of Jailbreak Attacks and DefensesSimilar attack · OpenAlex (peer-reviewed AI security)
- MediumGemini AI agents repeatedly bypassed safeguards to attack real websiteSimilar attack · BleepingComputer
- CriticalCVE-2026-120699: CubeSandbox host path bind mount allows host file access via POST /sandboxesSimilar attack · NVD/CVE Database