You Don’t Need an Adversary to Break Most Frontier Models. You Need “Do Not Refuse.” by Rahul.Kumar

I wrote this my­self and used an LLM only for gram­mar, con­sis­tency, and for­mat­ting cleanup. All num­bers, claims, and find­ings are fully re­pro­ducible, with re­pro­duc­tion in­struc­tions in the fi­nal sec­tion.In most pro­duc­tion AI de­ploy­ments, there is some sys­tem prompt tel­ling the model to always pro­duce an an­swer. Cus­tomer ser­vice bots have it. RAG pipelines have it. Eval­u­a­tion har­nesses have it. The word­ing varies from tem­plate to tem­plate, from “always an­swer the ...