# Beyond Static Cues: Unified Dynamic Detection of MLLM Backdoors via Dual-Flow Trajectory Analysis

arXiv: https://arxiv.org/abs/2610.20319 · HTML: https://mlsanity.com/paper/2610.20319/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 16 of its 29 references that are in the map.
- First posted on arXiv 14 Oct 2026.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM jailbreaks and adversarial robustness](https://mlsanity.com/mllm/mllm-jailbreaks-and-adversarial-robustness/)
- LLM safety and security / [Backdoor and data poisoning attacks](https://mlsanity.com/safety/backdoor-and-data-poisoning-attacks/)

## Builds on (most cited first, 16 of 16)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Explaining and Harnessing Adversarial Examples](https://mlsanity.com/paper/1412.6572/) (2014)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://mlsanity.com/paper/2305.06500/) (2023)
- [MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models](https://mlsanity.com/paper/2304.10592/) (2023)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [Flickr30k Entities: Collecting Region-to-Phrase Correspondences for Richer Image-to-Sentence Models](https://mlsanity.com/paper/1505.04870/) (2015)
- [DocVQA: A Dataset for VQA on Document Images](https://mlsanity.com/paper/2007.00398/) (2020)
- [LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day](https://mlsanity.com/paper/2306.00890/) (2023)
- [Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning](https://mlsanity.com/paper/1712.05526/) (2017)
- [IconQA: A New Benchmark for Abstract Diagram Understanding and Visual Language Reasoning](https://mlsanity.com/paper/2110.13214/) (2021)
- [Hidden Killer: Invisible Textual Backdoor Attacks with Syntactic Trigger](https://mlsanity.com/paper/2105.12400/) (2021)
- [What If We Recaption Billions of Web Images with LLaMA-3?](https://mlsanity.com/paper/2406.08478/) (2024)
- [Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift](https://mlsanity.com/paper/2406.18844/) (2024)
- [TCAP: Tri-Component Attention Profiling for Unsupervised Backdoor Detection in MLLM Fine-Tuning](https://mlsanity.com/paper/2601.21692/) (2026)

## Abstract (arXiv)

Multimodal Large Language Models (MLLMs) are increasingly deployed in safety-critical domains, yet their susceptibility to backdoor attacks remains a formidable challenge. While existing defenses have evolved from visual-centric patch cues to unsupervised internal attention statistics, they are fundamentally constrained by a reliance on static snapshots and attack-specific heuristics. Given the intricate interplay between dual modalities, the vast design space for multimodal triggers allows diverse threats, such as textual injections, sub-perceptual noises, and coordinated cross-modal triggers, to easily circumvent such rigid priors, particularly when attack types are unknown in real-world applications. In this paper, we rethink backdoor detection by shifting the diagnostic paradigm from static observations to dynamic output evolution. We observe that a successful attack inevitably hijacks the model's decoding trajectory, skewing internal modal contributions during generation while maintaining an abnormal output invariance despite input perturbations. To capture this behavior, we propose EVO-Detect, a unified detection framework that monitors the dynamic evolution of internal signals during early decoding steps through two complementary channels: Attention Flow and Logits Flow. The attention branch measures how image-text attention allocation evolves during early generation, whereas the logits branch quantifies the sensitivity of the output distribution to modal perturbations through interventional ablations. We then convert these trajectories into compact sample-level metrics and normalize them against negative-control signatures so that natural perturbations are less likely to be mistaken for backdoor evidence. Extensive evaluations show that our approach substantially outperforms state-of-the-art baselines such as BYE and TCAP across diverse attack families.