Metraipolis › archive › dc_655d313f0899

Render Before Reading: Visual Rendering as a Prompt Injection Defense — arXiv:2609.36121

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36121
License (as archived)
CC0-1.0
Archived
2026-10-07T16:25:21.062Z
SHA-256 of the archived text
4925c25a20eb2bbd2ca2bf06637c6396fa5ace02dee45e1e014f3debdc1298ef
Stable URL of this page
https://metraipolis.com/documento/dc_655d313f0899

How to cite

Render Before Reading: Visual Rendering as a Prompt Injection Defense — arXiv:2609.36121 — arXiv (en), https://arxiv.org/abs/2609.36121 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_655d313f0899 (Metraipolis archive dc_655d313f0899, sha256 4925c25a20eb2bbd2ca2bf06637c6396fa5ace02dee45e1e014f3debdc1298ef)
BibTeX
@misc{dc655d313f0899,
  title = {Render Before Reading: Visual Rendering as a Prompt Injection Defense — arXiv:2609.36121},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36121},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_655d313f0899; sha256 4925c25a20eb2bbd2ca2bf06637c6396fa5ace02dee45e1e014f3debdc1298ef},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36121. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models are vulnerable to prompt injection attacks, where third-party adversarial content can hijack the model's behavior. In this paper, we study the role played by the adversarial data's input modality, and identify a systematic asymmetry: multimodal LLMs are more likely to follow adversarial instruction when they appear as text than when the same instruction is delivered through a non-textual channel (e.g., as an image). We hypothesize that this modality gap arises from text-centric instruction tuning, which teaches models to obey textual instructions while treating other modalities mainly as content to parse or describe. We then demonstrate how this gap can be turned into a training-free defense, by rendering all untrusted payloads as typographic images (or audio) before they reach the model. Across ten models and two prompt injection benchmarks (DirectInject and AgentDojo) we show that our defense Pictionary consistently reduces attack success rates even against the strongest adaptive attacks and human red teamers, while largely preserving benign utility. We further show that benign fine-tuning on image-rendered instructions erodes the modality gap, tracing it to the text-centric instruction-tuning distribution.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.36121 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T15:54:53.777Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).