Vulnerability of LLMs to Vertically Aligned Text Manipulations