Reproduce “An In-Depth Investigation of Data Collection in LLM App Ecosystems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuhao Wu, Evin Jaff, Ke Yang, Ning Zhang, Umar Iqbal

Researchers from Washington University in St. Louis conducted an extensive empirical study of data collection practices within OpenAI's GPT app ecosystem, revealing that third-party Actions often collect broad ranges of user data, including prohibited types, with widespread omissions in privacy policy disclosures and weak platform enforcement. This investigation also found potential for indirect data exposure and cross-app user tracking due to shared execution contexts.

Key insights

  • Third-party Actions within the GPT ecosystem exhibit extensive data collection, often gathering sensitive and even explicitly prohibited information (e.g., security credentials, health data).
  • Privacy policies are largely deficient, with widespread omissions and vague disclosures regarding collected data, indicating a systemic lack of transparency.
  • Architectural design flaws, such as shared context windows, enable indirect data exposure between co-occurring Actions and facilitate cross-GPT user tracking.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.