OpenAI Restores Scheduled Tasks With Enhanced Scope Controls, Teases New Voice Model
View original source →OpenAI this week restored Scheduled Tasks functionality to ChatGPT—the feature allowing users to set recurring AI-executed workflows—after a technical and safety review, and separately confirmed a significantly upgraded voice model is in final preparation.
Scheduled Tasks restoration details:
• The feature was suspended in late May 2026 following reports of tasks executing with unexpected scope outside user-defined parameters • Restoration includes enhanced scope confirmation mechanisms requiring explicit user confirmation before first execution and each time parameters are modified • The confirmation UX surfaces the complete task definition and execution scope
Forthcoming voice model details:
• Described by internal sources as a 'groundbreaking' advance over current ChatGPT voice • Reported improvements in naturalistic speech quality, emotional range, real-time interruption handling, and multilingual capability • The model is separate from voice work in GPT-5.5, suggesting a dedicated voice-first architecture • Potential launch as early as the week of June 22
The dual announcements are part of a broader ChatGPT product expansion strategy aimed at demonstrating consumer AI product depth ahead of the IPO roadshow, where OpenAI will need to defend a $1 trillion-plus valuation against public market investors expecting product breadth beyond a chat interface.
The Scheduled Tasks scope confirmation mechanism should be studied by any organization building AI agents with recurring execution capability. The critical UX question for autonomous AI workflows is not 'can the user set it up?' but 'does the user understand and confirm what the agent will do, every time it does it?'
Why It Matters: The Scheduled Tasks restoration with enhanced scope confirmation demonstrates the correct design response to autonomous AI scope creep—addressing the problem at the user interface layer rather than restricting capability. A dedicated voice-first AI architecture would represent a qualitatively different product category.