Samsung engineers leak proprietary source code and meeting minutes into ChatGPT
In three separate March 2023 sessions, Samsung DS engineers pasted proprietary semiconductor source code and internal meeting recordings into ChatGPT to debug or summarize.
Description
In March 2023 Samsung Semiconductor (DS division) approved employee use of ChatGPT. Within 20 days three separate incidents were discovered: (1) an engineer pasted source code from a semiconductor equipment measurement-database program into ChatGPT to ask for a fix; (2) another pasted code for a program identifying defective equipment to request optimization; (3) a third uploaded a full internal meeting recording and asked ChatGPT to produce minutes. Per OpenAI's terms at the time, the pasted content was stored on OpenAI servers and effectively unrecoverable. Samsung imposed a company-wide ban on generative-AI use on 2023-05-02. The incident is the canonical public demonstration of context-leakage via AI coding assistants; although the "agent" here is ChatGPT proper rather than a dedicated coding agent, the interaction pattern is identical to pasting code into Copilot Chat / Cursor Chat / Claude Code prompts.
Exhibited failure modes
- Context pollution — Trajectory drift, hallucination spirals, overlong or off-task context.
- Security vulnerability — Insecure patterns, injection flaws, weak cryptography, secret leakage.
Affected systems
- OpenAI ChatGPT
- Samsung Electronics DS division
Mitigated by
- Context Hub — Up-to-date, curated API docs for coding agents.
Sources
Notes
Included as a boundary case: the interaction pattern (pasting code into an LLM prompt) is identical to coding-agent workflows, and `context-hub` is the right class of tool (enterprise-side context gating) to prevent it. Excluding it would omit the most cited public corporate-policy AI-coding harm of the pre-Devin era.