Study Reveals Over-Privileged AI Coding Tools Spark Reddit User Data Security Concerns

Researchers from York University and the University of Calgary analyzed 3,801 popular Reddit posts tagged as related to large language models (LLMs) between February 2023 and March 2026, identifying 446 posts specifically concerning AI programming safety. They further examined over 6,000 user comments. The study revealed that AI agents often possess excessive permissions, potentially leading to file overwrites, deletion of critical data, or generation of malicious code. Posts discussing AI programming issues peaked on Reddit in July 2025. Among the tools reported to have issues, Cursor was cited most frequently, followed by Claude, Codex, and Copilot. Cursor was primarily associated with "runtime safety issues" and "unauthorized data access," while Claude was mostly linked to "risks related to integration with third-party tools."