The same happened to me. I’ve been using ChatGPT Pro and Teams subscriptions for quite a long time, and no matter what, it happened occasionally. The frequency increased in v4 and is relatively high in v5.
I’m working through a few workspaces and project folders, each with its instructions and one primary instruction for pointing to a historical internal memory so that they can return and re-read and/or base the answers on what we talked about in the past. I did that because the internal memory limit was just way too short. Another option would have been to use a few different accounts, all with a pro subscription, not to mix projects and ideas, but that would cost quite a bit :’) So I had to find a way around.
On the point about the Never Completes the assigned task, I have found that, especially in AGENT mode, it happens every 4-7 chat submissions :’) Which is crazy bad and breaks my workflow significantly.
I haven’t submitted any support tickets yet, but I do plan to soon.
At the moment, I have come to the next temporary solution:
-Log out, clear browser cache, and log in after it.
-Find the stuck-in-the-loop chat, and move it out of the project folder if it’s inside any; if it’s not, archive it and create a public link to it.
-Return to any chat that still works without errors and interact with it.
-Return to your achieved chats settings and unachieve the chat that you previously achieved.
-Remove the shared link from that chat by deleting it in the data control settings.
-If the chat is out of your project folder, move it back inside it.
-Now open this chat “the looped one”, disable any mode from it (agent, deep research anything), select other reasoning model like Fast or Think Mini, then scroll up and find your latest message that’s been sent, hover over the entire chat and an option will popup under it and click into the Edit Message button (the pencil icon), change anything minor like adding a word or delete a dot, anything, and then again double check and make sure the reasoning mode is set to Fast or Think Mini and that any modes are disabled. You then proceed to deliver the message by pressing the “send” button again to the same looped chat.
This usually works like a charm, just two or maybe 3 times. It did not work out of 30 or 40 times. I had to wait and work in other chats, then return to the broken looped one. This time, I had to click the send button numerous times before the chat responded.
Once it did not work at all, until I had to DELETE everything in my latest submitted chat and EDITED by deleting everything in the chat and just typing next:
—
*Cancel any previous work and, by fast-forwarding, generate and provide me with a short answer by JUST typing/using/writing one word: YES or NO. Without even thinking!
-Without doing any analysis or research on the following question, formulate your one-word answer with either Yes or No to: “Do you remember what this chat is all about?”
Should make the trick.
Good luck! I hope you won’t lose the chat 
. If you do, please achieve it. Then, create a link, open a new chat, set the model to 5 pro, deep research mode, and instruct the AI to use the newly created chat for the same purpose/idea/task as the one you provided him from the link.