Source: to catch leakers, OpenAI security staff use a custom ChatGPT with access to Slack, email, and docs that cross-references news articles with access logs
Here at The Information, we're always trying to break through the hype to understand how folks are actually using AI in the real world.
Context & Ripple Effects
The reported internal tool follows disclosures that OpenAI’s internal messaging systems were breached in early 2023, exposing product details. That episode makes the company’s effort to join communications, documents, and access records a consequential shift in how it protects sensitive information the earlier internal-messaging breach.
It also sits alongside OpenAI’s prior work to build security controls into agentic systems, including red-team-informed safeguards such as watch-mode activation and terminal restrictions its agent security safeguards. The notable extension here is applying AI to internal investigations rather than only to product-facing safety.
First-order effects
- OpenAI security staff can use a single ChatGPT-based interface to correlate internal Slack, email, and document material with access logs and external reporting, potentially shortening leak investigations.
- Employees in the rollout markets face a more AI-mediated internal security process, with their work-system data becoming inputs to a tool designed to identify possible disclosure paths.
Second-order effects
- The deployment raises the operational bar for access controls, data retention, and investigation procedures: weak permissions or incomplete logs become more consequential when an AI system can rapidly connect them.
- It creates a concrete governance test for connected workplace AI: OpenAI must distinguish legitimate security review from overbroad monitoring while managing the sensitivity of the underlying communications and documents.
Third-order effects
- If this model spreads, enterprise AI governance will increasingly cover not only what assistants can generate, but also which internal systems they may query, how their inferences are reviewed, and how employees are notified or protected.
- The broader structural shift is toward AI as an investigative layer across the workplace stack, concentrating security analysis in tools that combine communications, knowledge stores, and identity or access data.
The trend: Connected AI agents are moving from productivity assistance into operational governance, where their value depends on controlled access to fragmented enterprise data.