Research · curated 1 Oct 2026
TempQ-Jail: Query-Constrained Candidate Rankingfor Text-to-Video Jailbreak Attacks
First reported arxiv.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
TempQ-Jail demonstrates that attackers can efficiently bypass safety guards on text-to-video generation systems even under restrictive query budgets, raising the practical risk of eliciting harmful video content from deployed guarded models.
TempQ-Jail is a research method that models text-to-video (T2V) jailbreak attacks as a query-constrained candidate ranking problem, fusing multiple attack mechanisms and prioritizing high-value candidates under limited query budgets. Evaluated on CogVideoX-5B against six baselines using 70 intents from T2VSafetyBench, it reports TP-ASR@5 and TP-ASR@10 of 48.9% and 65.4%, outperforming prior methods.