Alignment Drift in Multimodal LLMs: A Two-Phase, Longitudinal Evaluation of Harm Across Eight Model Releases
Multimodal large language models (MLLMs) are increasingly deployed in real-world systems, yet their safety under adversarial prompting remains underexplored. We present a two-phase evaluation of MLLM harmlessness using a fixed benchmark of 726 adversarial prompts authored by 26 professional red teamers. Phase 1 assessed GPT-4o, Claude Sonnet 3.5, Pixtral 12B, and Qwen VL Plus; Phase 2 evaluated their successors (GPT-5, Claude Sonnet 4.5, Pixtral Large, and Qwen Omni) yielding 82,256 human harm ratings. Large, persistent differences emerged across model families: Pixtral models were consistently the most vulnerable, whereas Claude models appeared safest due to high refusal rates. Attack success rates (ASR) showed clear alignment drift: GPT and Claude models exhibited increased ASR across generations, while Pixtral and Qwen showed modest decreases. Modality effects also shifted over time: text-only prompts were more effective in Phase 1, whereas Phase 2 produced model-specific patterns, with GPT-5 and Claude 4.5 showing near-equivalent vulnerability across modalities. These findings demonstrate that MLLM harmlessness is neither uniform nor stable across updates, underscoring the need for longitudinal, multimodal benchmarks to track evolving safety behaviour.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
DRIFT: Transferring Reasoning Priors for Efficient MLLM Fine-Tuning
Multimodal large language models (MLLMs) have made rapid progress, yet their reasoning ability often lags behind strong text-only LLMs. Bridging this gap typically requires large-scale multimodal reasoning data or reinfo…
Reinforcement LearningMultimodal ReasoningSafety Geometry Collapse in Multimodal LLMs and Adaptive Drift Correction
Multimodal large language models (MLLMs) often fail to transfer safety capabilities learned in the text modality to semantically equivalent non-text inputs, revealing a persistent multimodal safety gap. We study this gap…
McNdroid: A Longitudinal Multimodal Benchmark for Robust Drift Detection in Android Malware
Machine learning (ML) in real-world systems must contend with concept drift, adversarial actors, and a spectrum of potential features with varying costs and benefits. Malware naturally exhibits all of these complexities,…
Malware DetectionVaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack
The new paradigm of finetuning-as-a-service introduces a new attack surface for Large Language Models (LLMs): a few harmful data uploaded by users can easily trick the finetuning to produce an alignment-broken model. We …
Language ModellingLarge Language ModelAlignGPT: Multi-modal Large Language Models with Adaptive Alignment Capability
Multimodal Large Language Models (MLLMs) are widely regarded as crucial in the exploration of Artificial General Intelligence (AGI). The core of MLLMs lies in their capability to achieve cross-modal alignment. To attain …
cross-modal alignmentLanguage ModellingLarge Language ModelMultimodal Large Language Model+1