OpenAI is deploying a specialized version of ChatGPT to detect internal information leaks, according to The Information, citing a source familiar with the matter. When a news article about internal developments at OpenAI is published, security staff reportedly feed the text into this dedicated ChatGPT system, which has access to internal documents as well as employees’ Slack and email communications
The system can then suggest potential sources of the leak by identifying files or communication channels containing the disclosed information and determining who had access to them. It remains unclear whether OpenAI has successfully identified any individuals using this method.
OpenAI engineers recently presented the architecture of an internal AI agent that could be used for such purposes. The agent is designed to enable employees to perform complex data analysis through natural language queries and has access to institutional knowledge across Slack messages, Google Docs, and other internal systems.
AI Research Contributor
Daniel Mercer is an AI research contributor specializing in large language models, benchmarking, and multimodal systems. He writes about model capabilities, limitations, and real-world performance across leading AI assistants and platforms.
ES 
EN