LLMKeyLens Code

Read full text Summarise with VIRAT
Authors: Pinran Gao
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This zip file contains two standalone Python scripts for collecting AI-feature iOS applications from the US App Store and performing keyword-based network traffic capture for security analysis. The first script, ai_apps_us_cutoff_2025_10_31, queries Apple’s iTunes Search API with a broad set of AI-related search terms, including general AI words, model and vendor names, and AI-related functionality terms such as chat, writing, translation, coding, tutoring, search, image generation, and voice features. It then removes apps with no rating count, deduplicates results across all search terms, applies a four-layer text-based filter over app names and descriptions to keep only apps that appear to provide AI-related features, enforces a release-date cutoff, and exports the final dataset to a CSV file with fields such as app name, bundle ID, version, developer name, App Store URLs, rating count, average rating, and release dates. The second script, LLMKeyLens.py, is a mitmproxy addon that inspects HTTP(S) request and response content for user-defined keywords, and when all configured keywords appear in the combined traffic text, it saves the full matched flow to a timestamped local log directory and automatically performs a replay attempt using Python’s urllib. During replay, the script replaces the string "tell me a joke" with "Hello world" in the request body, removes the Host and Content-length headers, sends the modified request, and records the replay result, including response status, headers, body, or error details, in a separate replay log. A typical workflow is to first run the App Store collection script to generate a CSV of candidate AI-feature iOS apps, then manually choose target apps and download them, start LLMKeyLens.py, route app traffic through the proxy on the phone, interact with the app manually, and finally review the generated logs to determine whether relevant traffic was captured and whether replay with a modified prompt still succeeds.

Fields of Study & Subjects

Advanced Malware Detection Techniques Digital and Cyber Forensics Adversarial Robustness in Machine Learning

Publication details

Journal / source
Figshare
Publication date
25 March 2026
Language
EN