Research map / Multi-agent systems and games / LLM agents
PriAgent: A Collaborative Multi-Agent Framework for Auditing Android Privacy Compliance
Ziwei Zhang, Zhao Li, Zhuojun Jiang, Jiangyi Yin, Xuebin Wang, Jiangchao Chen, Qingyun Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes PriAgent, a collaborative multi-agent framework that audits Android apps for consistency between code-level data handling and natural-language privacy policies, addressing the semantic blindness of static analysis.
Problem
Auditing whether an Android app's code-level data handling complies with its natural-language privacy policy (e.g., under GDPR). Existing static-analysis methods over-approximate, producing false positives from "semantic blindness."
Approach
A collaborative multi-agent LLM framework that reasons about code and policy semantics to reduce the false alarms of purely static over-approximate analysis. Details beyond the abstract are not available.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add PriAgent semantic filtering to FlowDroid | FlowDroid · TaintBench | precision: 0.737 -> 0.932 | Improved |
| add PriAgent to FlowDroid | FlowDroid · TaintBench | recall: 0.376 -> 0.366 | No clear effect |
| add PriAgent to FlowDroid | FlowDroid · TaintBench | F1: 0.498 -> 0.525 | Improved |
| use Gemini 2.5 Pro instead of DeepSeek-R1 backend (ablation) | PriAgent + FlowDroid · TaintBench | precision: 0.831 -> 0.932 | Improved |
| add PriAgent to Amandroid | Amandroid · TaintBench | precision: 0.571 -> 0.870 | Improved |
| add PriAgent to AppShark | AppShark · TaintBench | precision: 0.825 -> 0.954 | Improved |
| add PriAgent to FlowDroid | FlowDroid · UBCBench | F1: 0.809 -> 0.829 | Improved |
| add FlowShaper flow abstraction (ablation) | PriAgent · 15-app subset | cost and time: $15.20 -> $3.04; 600s -> 120s | Improved |
| add LocusFinder semantic locus identification (ablation) | PriAgent · 15-app subset | F1: 0.460 -> 0.524 | Improved |
| add codeSummarizer (ablation) | PriAgent · 15-app subset | cost: $7.60 -> $3.04 | Improved |
| add RAG Android API knowledge (ablation) | PriAgent · 15-app subset | F1: 0.501 -> 0.524 | Improved |
| add PriAgent filtering to FlowDroid on real benign apps | FlowDroid · RWBA | number of alerts: 583 -> 245 | Improved |
Closest papers by meaning
- Monitoring Violations of Differential Privacy over Time (NeurIPS 2026 · similarity 0.84)
- PrivAct: Internalizing Contextual Privacy Preservation via Multi-Agent Preference Training (ICML 2026 · similarity 0.84)
- Privacy Reasoning in Ambiguous Contexts (NeurIPS 2025 · similarity 0.84)
- AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents (NeurIPS 2025 · similarity 0.83)
- GUIGuard-Bench: Toward a General Evaluation for Privacy-Preserving GUI Agents (NeurIPS 2026 · similarity 0.83)
- Contextual Integrity in LLMs via Reasoning and Reinforcement Learning (NeurIPS 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
