A new GitHub repository titled "Awesome Mobile On-Device AI Security" has been curated to serve as a roadmap for security research concerning AI models running locally on mobile devices.
The resource covers mobile AI runtimes including TFLite/LiteRT, Core ML, ExecuTorch, and ONNX. It details attack surfaces such as adversarial attacks, backdoors, model stealing, and energy-latency attacks. Defenses listed include model obfuscation, authorization, TEEs, and watermarking. The repository also provides a reading roadmap, taxonomy, open problems, and emerging directions.
The author invites feedback on missing papers, the clarity of the taxonomy, and the helpfulness of the minimal first-week reading path for beginners.