Hard to Read Is Not Hard to Audit: Why Degrading a Text Image Raises Jailbreak Success in One 2026 Study and Lowers It in Another, Four Accounts That Fit the Same Curve, and the Ablations That Would Separate Them

APA

Mahendrakar, P. (2026). Hard to Read Is Not Hard to Audit: Why Degrading a Text Image Raises Jailbreak Success in One 2026 Study and Lowers It in Another, Four Accounts That Fit the Same Curve, and the Ablations That Would Separate Them [Journal article]. Life of Research. https://doi.org/10.5281/zenodo.23427759

Copy APA

BibTeX

@misc{lor1122026,
  title        = {Hard to Read Is Not Hard to Audit: Why Degrading a Text Image Raises Jailbreak Success in One 2026 Study and Lowers It in Another, Four Accounts That Fit the Same Curve, and the Ablations That Would Separate Them},
  author       = {Pranay Mahendrakar},
  year         = {2026},
  publisher    = {Life of Research},
  doi          = {10.5281/zenodo.23427759},
  url          = {https://lifeofresearch.com/articles/112},
  keywords     = {multimodal large language models, jailbreak, typographic attacks, visual degradation, safety alignment, mismatched generalization, visual-text compression, attack success rate},
  note         = {LOR-00112}
}

Copy BibTeX More on the record page