Technology

Analysis details features and security implications of OpenAI's ChatGPT Work

An examination of the platform highlights cloud and local variants, internet-connected code execution, and potential prompt injection risks.

Updated Aug 31, 2026 · 02:18 UTC1 reports

The short version

OpenAI's ChatGPT Work product is divided into Work Cloud (web and mobile) and Work Local (desktop), restricted to subscribers paying $20 or more per month.
Work Cloud introduces features not available in standard Chat, including an internet-enabled code execution environment, persistent shared filesystem, headless Chrome browser, and website publishing capabilities.
The integration of private data access with exposure to untrusted web content raises unaddressed security concerns regarding potential prompt injection attacks.

OpenAI announced ChatGPT Work on July 9, 2026, offering it in Work Cloud and Work Local variants.[Hacker News]
ChatGPT Work is restricted to paid subscribers paying $20 per month or more.[Hacker News]
Work Cloud features internet-enabled code execution, allowing it to download GitHub repositories and interact with external domains by default.[Hacker News]
The platform includes a headless Chrome browser capable of executing JavaScript, taking screenshots, and navigating authentication flows without transmitting credentials to the model.[Hacker News]
Work Cloud allows users to deploy websites hosted via Cloudflare Workers and run sub-agent sessions using Sol, Luna, and Terra models.[Hacker News]

The exact security protocols OpenAI uses to protect ChatGPT Work sessions against prompt injection attacks remain undisclosed by the company.[Hacker News]