Forced Deferral: Manipulating Routing Decisions in Multimodal LLM Cascades
For developers deploying cost-efficient MLLM cascades, this work reveals a new security vulnerability where adversaries can manipulate compute allocation, leading to unintended resource consumption.
The paper introduces the Forced Deferral Attack (FDA), an adversarial image attack that manipulates confidence scores in MLLM cascades to force queries to be routed to the stronger, more expensive model. FDA achieves high success rates across datasets and model families, outperforming baselines in increasing strong-model routing.
While multimodal large language models (MLLMs) have shown strong visual reasoning abilities, serving a large model for every query is computationally expensive. MLLM cascades mitigate this cost by first querying a weak but cheaper model and deferring to a strong model when the weak model's output is unconfident. However, since the weak model's confidence directly controls compute allocation, these systems expose a new attack surface: an adversary can manipulate confidence so that their queries are consistently deferred to the strong model. Motivated by this vulnerability, we introduce the Forced Deferral Attack (FDA), an adversarial image attack that lowers the weak model's confidence and causes cascades to route queries to the strong model. FDA learns a universal border trigger by optimizing a temperature-flattened objective. This objective pushes the weak model's token distribution on triggered inputs toward less concentrated targets constructed from its clean responses. Across datasets, model families, and deferral metrics, FDA consistently increases strong-model routing while outperforming image-perturbation and prompt-injection baselines. These results show that MLLM cascades are vulnerable to attacks that manipulate compute allocation, forcing unintended strong-model usage without directly targeting answer correctness.