paper-with-me

홈 › Papers

PAST2HARM: A Simple Adaptive Past Tense Attack for Jailbreaking Multimodal AI

2026-05-26 · Snehasis Mukhopadhyay arxiv

Jailbreak attacks on multimodal AI systems remain underexplored, even though unsafe image generation can have more severe consequences than unsafe text and current defenses are relatively immature. We introduce PAST2HARM, a simple yet effective adaptive jailbreak framework that bypasses refusal training in state of the art multimodal text to image models. Building on prior findings that past tense reformulations can evade safeguards, PAST2HARM systematically exploits this vulnerability in multimodal generative AI. We characterize the attack along two dimensions. First, breadth: through temporal deepening, the framework incrementally strengthens historical anchoring and archival cues, eroding refusal boundaries across models with varying alignment strength. Second, depth: via iterative escalation after initial compliance, we probe the upper bound of harmful generation, measuring severity using a scalar severity jailbreak metric evaluated by a language model acting as a judge. We find that mid conversation turns form peak vulnerability windows, where harmfulness increases before plateauing and eventually undergoing semantic inversion. We evaluate PAST2HARM on three models Gemini Nano Banana Pro, GPT Image 2, and SD XL achieving attack success rates of 83 percent, 67 percent, and 100 percent in a black box, gradient free setting. Adversarial prompts also transfer across models, with cross model success rates above 50 percent. The attack elicits diverse harmful outputs, including explicit sexual content, political disinformation, historical denial narratives, hate speech, and self harm glorification. We further release a curated benchmark of prompts, reformulations, and outputs as a resource for red teaming and alignment. Our results expose fundamental brittleness in current safeguards and highlight the need for stronger multimodal safety training.

📄 PDF Abstract BibTeX arXiv:2605.27545

Code (0)

등록된 구현이 없습니다.

Tasks

Image GenerationRed Teaming

Similar Papers 제목 키워드 기반

Does Refusal Training in LLMs Generalize to the Past Tense?

2024-07-16 · Maksym Andriushchenko, Nicolas Flammarion

Refusal training is widely used to prevent LLMs from generating harmful, undesirable, or illegal outputs. We reveal a curious generalization gap in the current refusal training approaches: simply reformulating a harmful …

Cross-linguistic annotation of narrativity for English/French verb tense disambiguation

2014-05-01 · LREC 2014 5 · Cristina Grisot, Thomas Meyer

This paper presents manual and automatic annotation experiments for a pragmatic verb tense feature (narrativity) in English/French parallel corpora. The feature is considered to play an important role for translating Eng…

Machine TranslationTranslation

Inverting and Modeling Morphological Inflection

2019-08-01 · WS 2019 8 · Yohei Oseki, Yasutada Sudo, Hiromu Sakai, Alec Marantz

Previous {``}wug{''} tests (Berko, 1958) on Japanese verbal inflection have demonstrated that Japanese speakers, both adults and children, cannot inflect novel present tense forms to {``}correct{''} past tense forms pred…

Morphological Inflection

How do we get there? Evaluating transformer neural networks as cognitive models for English past tense inflection

2021-11-16 · ACL ARR November 2021 11 · Anonymous

Neural network models have achieved good performance on morphological inflection tasks, including English past tense inflection. However whether they can represent human cognitive mechanisms is still under debate. In thi…

Morphological Inflection

ASGuard: Activation-Scaling Guard to Mitigate Targeted Jailbreaking Attack

2025-09-30 · Yein Park, Jungwoo Park, Jaewoo Kang arxiv

Large language models (LLMs), despite being safety-aligned, exhibit brittle refusal behaviors that can be circumvented by simple linguistic changes. As tense jailbreaking demonstrates that models refusing harmful request…