ChatGPT: You can now interrupt long-running queries to refine what you're asking (help.openai.com)

OpenAI added a mid-query “update” that lets you interrupt long-running ChatGPT tasks (examples: Deep Research jobs or GPT‑5 Pro runs) to inject new constraints or facts without restarting or losing progress. From the sidebar you can click “update” and send information — e.g., “my wall is 72" wide and I can’t drill into it” — and the model will re-plan and continue the same job with the new requirements applied. This change matters because it tightens the human–AI loop for complex, multi-step workflows: users can correct assumptions, add constraints, or pivot goals mid-execution instead of aborting and repeating work, saving time and compute. Technically it implies ChatGPT preserves execution state and can recondition ongoing reasoning or agent plans on new input (dynamic context injection and incremental re-planning), which benefits research, design, coding and agentic tasks that often require iterative constraints. Expect smoother interactive debugging of long tasks, fewer wasted runs, and more practical human-in-the-loop control for advanced models like GPT‑5 Pro and Deep Research.