Researchtransformersjailbreakadversarial robustnessopenai moderation
SpatialJB Exploits Transformer Spatial Weakness To Bypass Guardrails
|
9.1
Authors Zhiyi Mou et al. submitted on Jan. 14, 2026 a paper introducing SpatialJB, a jailbreak method that redistributes tokens across rows, columns, and diagonals to exploit Transformer spatial weaknesses and disrupt LLM output generation. Experiments on leading models report nearly 100% attack success rates and sustain over 75% success against advanced filters including the OpenAI Moderation API; authors also evaluate baseline defenses and release code and a demo.
Scoring Rationale
High novelty, broad industry impact, and reproducible experiments with code; preprint status and limited external validation reduce verification certainty.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
Active Search Campaigns by BudgetEasyHigh CPC Clicks & Poor Landing PagesMediumCampaign ROAS by Attribution ModelHard
250 free problems · No credit card
See all Ad Tech problems
