Deep dive: The invisible privacy risks of using AI coding agents
kashifmanzoor · x · 2026-08-29
While using Codex and Claude for full-stack development, the author realized the risks extend far beyond source code leakage. AI agents read database rows containing real user data (names, emails, orders) during debugging, potentially sending this data to model providers without the user's explicit intent.
To mitigate this, the author adjusted their workflow:
- Switched the dev database to synthetic data instead of a production copy;
- Moved secrets out of the agent's read scope;
- Inspected query outputs for sensitive data before pasting;
- Carefully reviewed data retention and training terms between personal and enterprise plans.
The conclusion: when doing full-stack work with agents, the question isn't "is my code safe," but "what was in my context window this week."
More from coding & agent
- Andrew Ng: Why Software Engineering Fundamentals Remain Critical in the Age of AI Coding Agents — AndrewYNg · 2026-08-29
- LangChain Adds MCP Support in Open Source, Built on FastMCP — LangChain · 2026-08-29
- Dev built his own provider-agnostic artifact hosting after Claude's sharing limits — miihr_ · 2026-08-29
- Open-source 'universal pipe' connects cloud, local and self-hosted LLMs for free — conifer_v11 · 2026-08-29
- Better Models Need Good Design: 4 Levers for Coding Agents — rajistics · 2026-08-29
- LangChain Academy Hosting Live Workshop on Building Deep Agents — LangChain · 2026-08-29