@dazo @catsalad @jerry @Sempf @Viss @stux no but like (i assume) they have a layer between the actual LLM and the user-facing text, doing processing, and if it contains that string, replaces whatever response the LLM provided with a blocked message