ReCast protects private data in remote multimodal reasoning tasks
ReCast: Contract-Preserving Protection for Fixed-Interface Multimodal Reasoning
Cryptography and SecurityArtificial IntelligenceComputation and Language
Summary
Remote AI models that analyze multiple types of data, like charts and speech, can do complex reasoning but risk exposing private information when users send their inputs. The authors developed ReCast, a system that changes sensitive content locally into a protected form while keeping the important relationships intact for the AI to solve tasks. This method lets users keep data private while still benefiting from powerful remote AI reasoning. Tests show ReCast keeps accuracy high and reduces the chance that the AI sees sensitive details.
What this means in practice
- •For enterprise ai teams: Protect sensitive business data inputs while using third-party multimodal AI services for analytic tasks.
- •For cloud service providers: Offer privacy-preserving interfaces enabling clients to securely submit multimodal data without leaking confidential content.$Commercial implications: Enables privacy-focused AI processing services that appeal to security-conscious clients requiring guaranteed data confidentiality.
Authors
Bingchen Pei, Lichong Chen, Bingxi Zhao, Ziang Wu, Sirui Wang, Min Zhang, Yanhao Chen, Qingxu Liu, Qiang Gao, Chang-Tien Lu, Bo Gao
Abstract
Remote multimodal models offer strong numerical reasoning capabilities over charts and speech, but sending private inputs risks exposing sensitive content. Text-only sanitization cannot directly satisfy fixed media interfaces, while identity anonymization leaves the underlying task content exposed. We introduce ReCast, an agentic plug-in framework that replaces source-specific content while preserving task-relevant relations and the required input modality. ReCast locally converts inputs into a shared textual evidence-query record, jointly rewrites entities and topics with a distilled 4B model, and substitutes values through a locally invertible, role-aware numerical map. A reconstruction agent generates and validates the required media from the protected record. The remote solver returns a program whose protected operands are restored locally before execution. On 4,000 held-out ChartQA and NMSQA examples, ReCast achieves 75.10% accuracy, retaining 92.43% of unprotected remote accuracy, while a model-based audit flags source-content leakage in 7.95% of solver-bound requests. It outperforms all evaluated local baselines, preserving the benefit of remote reasoning while reducing source-content exposure under existing media interfaces.