Prompt injection in Google Translate can reveal the underlying instruction-following language model. Responses indicate that the model lacks strong boundaries between processing content and following instructions.