AI RESEARCH
Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs
arXiv CS.LG
•
ArXi:2604.10403v1 Announce Type: new We address jailbreaks, backdoors, and unlearning for large language models (LLMs). Unlike prior work, which trains LLMs based on their actions when given malign instructions, our method specifically trains the model to change how it interprets instructions. Our method, Latent Instruction Representation Alignment (LIRA), greatly improves generalization. We further boost generalization through an internally adversarial