Research · curated 29 Jul 2026
Jailbreaking Large Language Models via Multi-Task Embedding-based Prompt | Proceedings of the 2026 IEEE/ACM Third International Conference on AI Foundation Models and Software Engineering
First reported · updated · 2 reports acm.org
Coverage timeline
Why it matters
MTEA demonstrates that exploiting an LLM's degraded performance under concurrent multi-task load can reliably bypass safety alignment and existing jailbreak defenses on production models like GPT-4o and Gemini, raising the bar for defenders.
Researchers present the Multi-Task Embedding-based Attack (MTEA), a jailbreak technique that embeds malicious instructions within three concurrent tasks (Code Understanding, Language Translation, and Pattern Adherence) to disrupt LLM safety alignment. Evaluated on six models including GPT-4o and Gemini-2.5-pro using AdvBench, MTEA reportedly achieves 100% attack success and following rates, defeats Perplexity Filter and SmoothLLM defenses, and reduces query costs by 90% versus baselines.