mlsanityData to 7 Oct 2026

Mind Your Weight(s): A Large-scale Study on Insufficient Machine Learning Model Protection in Mobile Apps

arXiv 2002.07687 ↗First posted 18 Feb 2020cs.CR cs.LGcited by 35

Not on the map. Why?

Abstract (arXiv)

On-device machine learning (ML) is quickly gaining popularity among mobile apps. It allows offline model inference while preserving user privacy. However, ML models, considered as core intellectual properties of model owners, are now stored on billions of untrusted devices and subject to potential thefts. Leaked models can cause both severe financial loss and security consequences. This paper presents the first empirical study of ML model protection on mobile devices. Our study aims to answer three open questions with quantitative evidence: How widely is model protection used in apps? How robust are existing model protection techniques? What impacts can (stolen) models incur? To that end, we built a simple app analysis pipeline and analyzed 46,753 popular apps collected from the US and Chinese app markets. We identified 1,468 ML apps spanning all popular app categories. We found that, alarmingly, 41% of ML apps do not protect their models at all, which can be trivially stolen from app packages. Even for those apps that use model protection or encryption, we were able to extract the models from 66% of them via unsophisticated dynamic analysis techniques. The extracted models are mostly commercial products and used for face recognition, liveness detection, ID/bank card recognition, and malware detection. We quantitatively estimated the potential financial and security impact of a leaked model, which can amount to millions of dollars for different stakeholders. Our study reveals that on-device models are currently at high risk of being leaked; attackers are highly motivated to steal such models. Drawn from our large-scale study, we report our insights into this emerging security problem and discuss the technical challenges, hoping to inspire future research on robust and practical model protection for mobile devices.

Key facts. Cited by 35 papers in the map (data through 7 Oct 2026). First posted on arXiv 18 Feb 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 3; 2022: 7; 2023: 9; 2024: 7; 2025: 6; 2026: 3.

35 papers.

Sort:

By area:

  1. 2022"Real Attackers Don't Compute Gradients": Bridging the Gap Between Adversarial ML Research and Practicecited by 55
  2. 2021Smart at what cost? Characterising Mobile Deep Neural Networks in the wildcited by 25
  3. 2023No Privacy Left Outside: On the (In-)Security of TEE-Shielded DNN Partition for On-Device MLcited by 21
  4. 2021Robustness of on-device Models: Adversarial Attack to Deep Learning Models on Android Appscited by 19
  5. 2023NNSplitter: An Active Defense Solution for DNN Model via Automated Weight Obfuscationcited by 15
  6. 2024SLIP: Securing LLMs IP Using Weights Decompositioncited by 13
  7. 2022Smart App Attack: Hacking Deep Learning Models in Android Appscited by 12
  8. 2024Octopus v4: Graph of language modelscited by 11
  9. 2023Efficient Query-Based Attack against ML-Based Android Malware Detection under Zero Knowledge Settingcited by 10
  10. 2021Confidential Machine Learning Computation in Untrusted Environments: A Systems Security Perspectivecited by 4
  11. 2023A Light-weight Deep Learning Model for Remote Sensing Image Classificationcited by 4
  12. 2022Edge Security: Challenges and Issuescited by 3
23 more
  1. 2022A Robust and Low Complexity Deep Learning Model for Remote Sensing Image Classificationcited by 3
  2. 2023InferNet: Exploiting Aggregate GPU Profiles as Side-Channel for DNN Architecture Inferencecited by 3
  3. 2024TEESlice: Protecting Sensitive Neural Network Models in Trusted Execution Environments When Attackers have Pre-Trained Modelscited by 3
  4. 2024Towards Data Governance of Frontier AI Modelscited by 3
  5. 2022DeepTaster: Adversarial Perturbation-Based Fingerprinting to Identify Proprietary Dataset Use in Deep Neural Networkscited by 2
  6. 2023Beyond the Model: Data Pre-processing Attack to Deep Learning Models in Android Appscited by 2
  7. 2023Neural Dehydration: Effective Erasure of Black-box Watermarks from DNNs with Limited Datacited by 2
  8. 2024A Picture is Worth 500 Labels: A Case Study of Demographic Disparities in Local Machine Learning Models for Instagram and TikTokcited by 2
  9. 2024TBNet: A Neural Architectural Defense Framework Facilitating DNN Model Protection in Trusted Execution Environmentscited by 2
  10. 2025THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Modelscited by 2
  11. 2022Robust, General, and Low Complexity Acoustic Scene Classification Systems and An Effective Visualization for Presenting a Sound Scene Contextcited by 1
  12. 2022An Empirical Study of AI Techniques in Mobile Applicationscited by 1
  13. 2025Stealthy Backdoor Attack to Real-world Models in Android Appscited by 1
  14. 2025"Impressively Scary:" Exploring User Perceptions and Reactions to Unraveling Machine Learning Models in Social Media Applicationscited by 1
  15. 2025WGLE:Backdoor-free and Multi-bit Black-box Watermarking for Graph Neural Networkscited by 1
  16. 2023Evil from Within: Machine Learning Backdoors through Hardware Trojanscited by 0
  17. 2023Towards Real Smart Apps: Investigating Human-AI Interactions in Smartphone On-Device AI Appscited by 0
  18. 2024A Novel Access Control and Privacy-Enhancing Approach for Models in Edge Computingcited by 0
  19. 2025ProDiF: Protecting Domain-Invariant Features to Secure Pre-Trained Models Against Extractioncited by 0
  20. 2025NeuroDeX: Unlocking Diverse Support in Decompiling Deep Neural Network Executablescited by 0
  21. 2026LoREnc: Low-Rank Encryption for Securing Foundation Models and LoRA Adapterscited by 0
  22. 2026SoK: Attack and Defense Landscape of Mobile On-device AI Systemscited by 0
  23. 2026Privacy-Friendly Cohort Determination: Sealed, CSP-Independent In-Browser ML Inference of Professional Segments for Identity-Less Advertisingcited by 0