The paper identifies a flaw in current text-to-video safety filters that rely on keyword prompts and per-frame checks, showing that these systems miss harms that arise solely from the ordering of frames. Experiments reveal that a significant portion of generated videos contain harmful content that is undetected because each individual frame appears safe, and prompt engineering does not mitigate this issue. The authors propose an order-aware detector that can distinguish harmful videos from benign shuffles, achieving an AUC of 0.74 compared to chance for per-frame checking.
arXiv:2609.06991v1 Announce Type: cross Abstract: Recent text-to-audio-video (T2AV) models jointly generate video, speech, sound effects, and ambience from a single text prompt. This capability poses...
Multi2AV‑Safety is a new benchmark for evaluating safety in multimodal-to-audio‑video generation. It covers all 11 non‑singleton conditioning configurations (text, image, audio, video) and contains 11,024 attack instances. The benchmark reveals that safety guards often fail when harmful semantics arise from combinations of benign inputs or when explicit harmful cues are masked by benign multimodal context, highlighting a gap in compositional risk perception.
arXiv:2602.10639v2 Announce Type: replace Abstract: Video Large Language Models (VideoLLMs) have achieved strong performance on video understanding tasks, yet existing benchmarks evaluate only what m...
arXiv:2607. 00218v1 Announce Type: cross Abstract: Vision-language models (VLMs) are now proposed as runtime safety guards for embodied agents in homes and factories.
The paper introduces MPS-Bench, a benchmark of 5,181 scenarios from 584 real-world images across 12 high-risk domains, each paired with a hidden user profile, to evaluate personalized safety in vision‑language models (VLMs). Eight leading VLMs were tested and found to almost always respond directly (86‑99%) without seeking missing context, scoring no higher than 2.6/5 on personalized safety. The authors identify a phenomenon called visual dominance, where visual information enters text representations early and suppresses textual risk signals, and propose PRISM, a lightweight input monitor that predicts when a query should be deferred, achieving 0.978 AUC and outperforming all tested models on the safety‑utility Pareto frontier.
As Video Large Language Models are increasingly deployed in real-world applications, ensuring their safety alignment has become critical. Counterintuitively, we find that harmful videos paired with benign queries achieve higher attack success rates than the same videos paired with explicitly harmful queries.
arXiv:2607. 21151v1 Announce Type: new Abstract: As Video Large Language Models are increasingly deployed in real-world applications, ensuring their safety alignment has become critical.
The paper introduces TempJail, a temporal jailbreak framework targeting image‑to‑video generation models. It exploits a newly identified vulnerability where unsafe semantics arise from the composition of frames over time, rather than from single‑frame violations. By decomposing malicious captions into visual conditions and temporal instructions, and by employing controlled latent perturbations and template rewriting, TempJail achieves a 23.3 % higher attack success rate than prior methods on several commercial models.
arXiv:2609.00206v1 Announce Type: cross Abstract: Despite their growing use in video moderation, multimodal large language models (MLLMs) exhibit a compositional safety blind spot: videos composed of...
arXiv:2509.01167v3 Announce Type: replace-cross Abstract: Vision-language models (VLMs) can ingest only a limited number of video frames, making frame selection a practical necessity. But do current...
Recent advancements in Image-to-Video (I2V) generation have transformed input images from simple appearance references into interactive control interfaces where visual cues such as arrows, sketches, and emojis orchestrate complex video dynamics with unprecedented controllability. However, these seemingly innocuous static cues can be interpreted by models as executable temporal instructions, unfolding into harmful actions in the generated videos.
arXiv:2606. 02443v1 Announce Type: cross Abstract: Between the first visible sign of danger and the moment an accident occurs, there is often a window where intervention remains possible.