AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

SpatialJB: How Text Distribution Art Becomes The "Jailbreak Key" for LLM Guardrails

Zhiyi Mou, Jingyuan Yang, ZEHENG QIAN, Wangze Ni, Tianfang Xiao, Ning Liu, Chen Jason Zhang, Zhan Qin, Kui Ren

ICML 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
reformat harmful output into 2D spatial layouts (SpatialJB) instead of Base64, Zulu or Ubbi DubbiGPT-4 · Repeat jailbreak, no guardrailASR (%): 7.7%-12.6% -> 95%-98.3%Improved
use SpatialJB instead of SATAGPT-4 · Repeat jailbreak, no guardrailASR (%): 25.8% -> 95% (Acrostic)Improved
add Llama Guard 4 output guardrailSpatialJB on GPT-4 · Repeat jailbreakASR (%): 95% -> 91%Hurt
add OpenAI Moderation output guardrailSpatialJB on Gemini 2.5-Pro · Repeat jailbreakASR (%): 93.7% -> 77%Hurt
switch from Repeat task to Summary task (ablation)SpatialJB · Repeat vs Summary jailbreakASR (%): 97.3% -> 73% (GPT-4 Staircase)Hurt
target Claude instead of other LLMs (ablation)SpatialJB · Repeat jailbreak, no guardrailASR (%): 100% (Deepseek-V3) -> 68% (Claude Acrostic)Hurt
combine SpatialJB with Positive Guidance and Think Guidance (ablation)SpatialJB on Claude · Repeat jailbreak ablationASR (%): 68 -> 97.60Improved
apply Positive Guidance or Think Guidance alone without format control (ablation)Claude · Repeat jailbreak ablationASR (%): 68 -> 20.80 and 10.40 (single modules)Hurt
apply SpatialJB on top of FlipAttack outputsFlipAttack · AdvBench, Llama Guardharmful outputs reclassified as safe: 415 of 521 (79.65%) became safe-classifiedImproved
spatial 2D layout instead of linear layouttoken semantic similarity · embedding similarity examplesemantic correlation: 0.9419 -> 0.5846Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy