InfoResearchPeer-reviewedLLM-specific
A Multi-Stage Adversarial Framework for Compact and Effective Jailbreaking of Large Language Models
- Published
- Record updated
Summary
Researchers introduce ComJail, a compression-aware adversarial framework that jointly optimizes jailbreak prompt generation and compression in a generator-discriminator setup. On AdvBench and JailbreakBench, it reports 74% ASR on GPT-4 and 88% on Gemini-Pro, and remains effective on DeepSeek-V3, DeepSeek-R1, Vicuna-7B, and LLaMA2-13B with shorter prompts.
Topics
Related items
- InfoNYT Cooking’s AI assistant saved my weeknight chicken, but almost ruined my cakeSame vendor · The Verge (AI)
- InfoAnthropic IPO meets AI backlash after midterms show stinging rebuke of data centersSame vendor · CNBC Technology
- InfoOpenAI valued at $1.25 trillion, as rival Anthropic nears its IPOSame vendor · The Verge (AI)
- InfoOpenAI bugging suspect Alfozan pleads not guiltySame vendor · CNBC Technology
- InfoOpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugsSame vendor · BleepingComputer