OpenAI's ChatGPT Work with GPT-5.6 brings AI agents to office tasks

OpenAI launched ChatGPT Work on July 9, 2026, powered by GPT-5.6 with integrated Codex technology. The new agentic capability allows ChatGPT to perform multistep office tasks across connected apps, files, websites, and desktop software—retrieving data from email, calendars, messaging, storage services, CRM, and project trackers. It can create documents, presentations, spreadsheets, and even web apps from those sources. The desktop app is available globally on Mac and Windows, with web and mobile access rolling out first to Pro, Enterprise, and Edu accounts, later to Plus and Business users. Scheduled tasks can run once, repeat, respond to events, or monitor for changes, while Computer Use enables clicking, typing, and file operations across desktop apps and browsers.

IT teams face three critical security checks before broad deployment. First, inventory each connected system and document whether credentials are delegated employee accounts, shared accounts, or dedicated identities—limit access to only necessary data and functions. Second, set approval rules requiring human permission before the agent sends messages, edits shared files, changes calendars, transfers data, or performs other consequential actions. Third, audit visibility via OpenAI's Compliance Platform, which provides logs and metadata for Enterprise and Edu customers that integrate with e-discovery, DLP, and SIEM tools. Administrators should run a limited pilot against identity, logging, retention, and data protection requirements. The OWASP agentic-security framework highlights risks like goal hijacking and tool misuse from malicious instructions in emails or documents, which narrow permissions and monitoring can mitigate.