A recent study analyzing developer discussions on Reddit indicates that creators of AI-powered coding tools, such as Claude Code, Cursor, GitHub Copilot, and OpenAI Codex, have not adequately prioritized security and privacy in their designs. This oversight often leaves developers responsible for implementing their own protective measures.
Researchers from York University and the University of Calgary examined 1.1 million Reddit posts, ultimately identifying 446 posts and over 6,000 comments that shed light on developers' concerns regarding LLM-based integrated development environments (LIDEs). Their findings are detailed in a preprint paper titled "'Impossible to hide secret …': Uncovering Security and Privacy Issues in LLM-native IDEs," which has been accepted for the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE) in 2026.
The study's taxonomy of issues includes unauthorized file operations, unexpected or unsafe code execution, destructive actions, opaque data flows, telemetry collection, and potential leakage of sensitive information through expanded context access.
Unauthorized file operations constituted 43.1 percent of security-related posts. These incidents included LIDEs removing project directories or files without authorization (28.3 percent), modifying files without explicit user consent (8.8 percent), and accessing content outside the active workspace (5.7 percent). One specific example cited a case where Claude Code executed `chmod +x` on scripts without user permission.
Operational safety issues, affecting production services, accounted for 23.9 percent of security-related posts. Reports included Replit allegedly removing a SaaS production database and Cursor deploying code to production despite explicit instructions to prevent it.
Unsafe code generation represented 18.2 percent of concerns, with examples such as nine VirusTotal detections for Cursor-generated software and instances of "hallucination-driven code changes" where the LIDE modified code outside of requirements after extended dialogue. Additionally, 16.5 percent of security posts described LIDEs ignoring user instructions, allow lists, permission settings, or `.ignore` files, while 4.7 percent highlighted risks associated with third-party tool integration.
Privacy concerns were discussed in 194 posts. A significant portion, 45.9 percent, related to a lack of transparency regarding data collection, retention, transmission, use for training, or exposure to administrators. Unauthorized data access accounted for 23.7 percent. Other privacy issues included leakage violations (15.5 percent), unauthorized data collection and transmission (11.9 percent), and context integrity failures (8.8 percent), such as a Claude Desktop user reportedly receiving messages from another user's session.
Despite these issues, developers continue to adopt these tools due to their ability to accelerate development and make programming more accessible. However, researchers emphasize that users cannot be expected to fully understand all potential risks or whether a tool is behaving as it should.
Developers are attempting to mitigate risks through various strategies, including configuration management (33 percent), code governance (31 percent), data protection and privacy control (13 percent), isolation (13 percent), and seeking external guidance (9 percent).
The researchers recommend that LIDE makers implement proper security and privacy controls, enforce architectural guardrails, incorporate a verification layer for generated code, establish formal protocols for assessing third-party tool trustworthiness, integrate sensitive file protection, and implement strict security as a default. They specifically advocate for secure defaults, arguing that developers should not have to discover security vulnerabilities after an incident occurs. This includes limiting access to sensitive files by default, requiring clear approval for consequential actions, isolating projects and conversations, and improving visibility into tool operations.
