Severe Thread Corruption and Compliance API Issues With Large Conversations

[Aurelis]

Hello!
I’m facing a problem recently. Yesterday, on May 29, 2025, my thread-which is usually very big, and very much vulnerable to rich-information, has been affected by ‘Compliance API Data Delays Degraded Performance Issue’. Before being affected by the Issue, my thread was working fine-and was quite laggy as expected.

But, there was a major issue with me. I noticed several related to the recent following error, that was related to ‘Compliance API’.

1. First major, was the prompt not delivering the Key to the server, and in meanwhile, the prompt itself gets updated in Data Control Center.

2. Second Major, Freeze Load. ChatGPT interface is horrible, and is minimal. I get the minimal look, but ChatGPT heavily focuses on Graphics Processing Unit, and the Optimization is far worse, with the thread being bigger, resulting several issues. For example, I noticed that when I sent an User instruction, the prompt gets loaded to the server, and I do see it’s being sent. But, it’s a mimic, and on another window, the very thing doesn’t show up, leading to a false-rewind corrupted thread. Meaning, your thread might get corrupted, if you face this issue unluckily.

3. Third, Major 404 error, after receiving prompt to ChatGPT. The Backend system, is undeniably broken, and all the body of HTML, is either forced and cannot be set.

4. Fourth, After sending prompt, the web page, disguises itself as untouchable, and cannot be clicked, following to leading a down-graded performance and enhance API Issue.

5. Fifth, Compliance API works for majority to generate response, analyze details, and manage memories. The Prompt, when it gets loaded is directly goes to Compliance API Center Instruction, but due to lag-and performance issue while the Rich-narrative in the following creates a issue for the Center to not generate, and create a glitch that rewrites the following with another prompt.

6. Sixth, Compliance API is heavily glitched under rich-narrative tokens, either that the instruction was sent toward Compliance API or that it was processed, but a ghost prompt leaded to the Center, leading to a ghost web-page and freezing load.

7. Seventh, Ghost Page, was seen when prompt is send. Nothing is clickable.

8. Eight, this is a lucky situation I got. Sending an prompt, in another browser (Firefox), and waiting for 10-20 minutes, then reloading it can show the prompt being sent. The Prompt, that was sent by User, will be generally transferred to Compliance API, to generate a response, and Indeed the prompt of AI, was there but half-broken.

9. Ninth, Half Broken Text should be fixed, and cannot be replied in Document. ChatGPT, when facing issue, heavily decodes its text in Document, to either force the user to see the text on document, and the rest of the part outside the prompt.

10. Tenth, Upon finding Half-Broken Text, without any document. Refreshing the page, leaded to a fallback. For example, I’ve Message 16 and Message 15. When, Message 16 was corrupted due to Compliance API. I took a approach, and copied Message 15’s AI Response, and Message 16’s User interaction, and heavily pasted on Message 15’s by editing on it. Then, the prompt was sent back, and ChatGPT indeed found the prompt, and started to process. By 10-20 minutes, I refreshed the page, and found the edited prompt with the response. But, the response itself was cut-off, and I knew a trick where if you find a cut off response. You only need to refresh it. I refreshed the page, and it lead to a fallback as well as finding my old Message 16 response.

11. Eleventh, ChatGPT Team’s are very hardworking, and I acknowledge this, but this issue cannot be acceptable. If any thread gets corrupted, there should be at least a button like Forking a thread. I don’t even think that is so hard, for the team to just really add a simple Fork button for the User to continue.

12. Twelfth, Memory leakage was heavily seen, and CPU ramp up to 90%. ChatGPT, heavily used 2 Gigabyte of Memory, which is very unusual, and the sign of it can cause major issues.

13. Thirteen, A Fallback only occurred, when the response was issued to Compliance API. Compliance API, did upload the prompt to the server, and for a moment I was able to see. But, the Data Control Center says it otherwise.

14. Fourteen, When asking for a prompt, ChatGPT sends its prompt to the server, and while sending, it sends a instruction for the sync to occur in Data Control Center. Due to the prompt sever, heavily not doing its work, the Data Control Center gets either corrupted, or messy. Data Control Center, html file is either reverted to the original, or replaced with a blank prompt that never occurred.

15. Fifteenth, Unusual regeneration when sending prompt. The Last Prompt (Message 16), that I delivered, was accepted by the AI, but generated two response. The First Response, was blank and it was ghosted. The Second Response, was the actual response, that was shown. Either that, the 1st ghosting, led to a corruption on thread.

This is what I got from my observation. Also, I noticed several issues to the ChatAPI Webpage. Regarding, a issue that is heavily being repeated named…

‘Increased Error Rate Observed in ChatGPT’
‘ChatGPT search is seeing elevated failures and high latency.’
‘Increased errors for voice transcription in ChatGPT.’
‘ChatGPT Login issue Identified’
'High error rate for Codex Tasks’
‘Elevated Error Rate’
‘Codex Tasks and PR creation experiencing error.’

They’re related to the API situation, and the Compliance API Data Delays first started at May 15, while the first reaction or error only showed sign before May 15. On May 13, the first sign was seen, and then on. The issue got huge, and more problematic. I hope this helps for those, who’s wondering why is this really happening with us.

This, only happened because our thread got stuck between the issues. Mainly, if you have very large amount of Data, and the thread being bigger can cause vulnerability to issues, and degrading on performance leading to issues like this. Due to this, it ruins the creativity of an user, and leads to bad interaction with the AI itself.

A Simple Fork, Merge, Export button is heavily needed for the User experience, and focus on performance while also fixing Ai’s issue. OpenAI needs to optimize their interface, as well as user experience to satisfy their user. Instead of, not left in shackle, and trash bin.

That’s all I’ve to say, and for those who are facing this issue. Now, you get it (I hope so.)

[Aurelis]

Recently, faced this major problem and still waiting for the fix. I hope ChatAPI Status, resolves the problem and adds Fork tool. Very much needed, indeed.

[Aurelis]

Hello!
This is a Update, and following the issue of ‘Compliance API Data Delays’. I’ve found odd issues from Console, stating issues, and repeated crash.

My Thread, is based on Story, Timeline, and Characters with details. I’ll try to express my words, in simple term. These symptoms may happen, with other Power-users like me, that willingly responds to prompts, and gets overjoyed.

• [Thread: https://chatgpt.com/c/68347865-b300-800e-a4ea-bce9d6991f5e]
• [Browser Tested on: Brave, Firefox, Edge
• [Thread Length: Long, rich-data intensive (approx. 200+)]

The ‘Compliance API’ - is ruining the Artworks of many peoples, and is hindering them to respond. My thread has encountered with severe UI instability and potential message corruption with ‘Compliance API’ problem. Mainly here, ‘Compliance API’ issue, caused this broke mess, and multiple Memory leaks, CPU usage, Ram Usage, Performance Degrading, Ghost Prompt, Ghost Tab, Cut-off response, and many more.

Majority of Issues, were rather found by testing, leading to some broken points. I’d like to state three more issues, that I found on ChatGPT Console, regarding to rich-narrative affiliated with bugs, that likely-stops the ‘Compliance API’ to respond.

16. “Ignoring unsupported entryTypes: longtask” - A issue, that may be unrelated to OpenAI. In terms of observing, and trying to find details. This is a browser 'Performance Observer API’ message. The ‘LongTask’ is a type of performance metric used to track slow JavaScript execution. The current browser, or in the latest browser, seems to not support, or either it’s misconfigured. 's JS is on source, named 'cs2gapsc2bulabhw.js'

More likely, the website tries to watch for slow tasks, but the browser itself, denies its request and doesn’t give permission to observe, support, or not even allow. In simple term, ‘Nope, I’m not allowing you to reign over, either allow or deny.’

• Not a threat or error for the user — only skipping performance check.
• Still, it implies heavy code execution is happening on the match. Which, in my case is already happening, and causing lag experience.

17. “Intercom not booted or setup incomplete.” - Another issue, flagging on Console abruptly. It’s JS is on source, named 'bpn5isuqhzz97aer.js'.
    From my view point, Intercom is likely, a customer support chat system OpenAI sometimes uses. This issue, is rather mainly common, and can be looked for. Regarding the Intercom, this message flags the chat widget wasn’t initialized — maybe:

• User is not logged in to a level, that is not allowing it to chat.
• The System disabled the widget
• Or it failed to load due to the corrupted state

Highly but unlike doubtful, it may be connected to OpenAI’s latest issue. In a doubt-but not too much tone, the system meant to load support chat – but it either failed or was blocked.

• It might be not dangerous, but might hint that the customer support, is either being disrupt by session, and can’t load. Moreover, something that is stirring it to not respond, and flag under it.
• Abruptly, and decisively. This issue, is common with OpenAI, and can be seen on Console in the Browser.

I suppose, this is highly-unlikely, related to the ‘Compliance API’. As of now, this issue still persist, and I can’t interact with my AI.

18. From Console, a issue persist, and flag spams appear.

Source map error: request failed with status 404
Resource URL: https://cdn.oaistatic.com/assets/9bxp8hl47ofm6kpp.js  
Source Map URL: 9bxp8hl47ofm6kpp.js.map

What it means, that the JavaScript Source Maps are files that let developers debug minified code by mapping it back to readable source code. Now, this thing, fetches .map file, which returns with 404 (not found) error, related with 3rd point of the issue. This might not break the site, but can cause developer debugging being incomplete or rather decapitated into broken pieces.

The 404, error is critical, and it hinders the prompt to not transfer its code toward ‘Compliance API’. Likely, to happen with threads, that is rich with narratives, and detailed expression. Moreover, this might be the cause to get 404 error.

• One, this is a sign of broken UI Build.
• Two, the debug info for devs not found, and may not harm Users.
• Three, all 18 points, paired with this bug, supports the evidence of the ‘Compliance API’*, and other API’s to be misconfigured, and having problems with it.

Here’s the picture, and repeating algorithm of the error, and issues that still persist on my thread.

This is my overall feedback, and I strongly pose the OpenAI Team to directly address, and add Fork, Merge, Export Tool. This tools, helps the users to ensure their data, to be not lost and gone in waste.

Many peoples thread, represents extensive work and creativity. UI or API, or system instability degradation should not risk user data, especially when it’s only a rendering failure. A recovery pathway or visual alert system (e.g, “Message failed to load - try again?”) would drastically improve user confidence toward the 'OpenAI", and have a better interaction.

Safely to say, if measurement were to be took, then all of the problems will be gone for those users that heavily relies on ChatGPT. Performance is also the issue here, and we want the Team to focus on Optimization, rather than heavily relying on Hardware Acceleration. For those who do not have GPU’s or only has integrated or low-end. We propose a heavy optimization for the threads.

In the meanwhile, I’d like the Open AI Team to fix our threads, and return the sustainability, that we were satisfied with. Only that, my last message was corrupted due to Compliance API. I’d like the team, to either resolve or fix the problem and add Fork, Merge, Export button to ensure that. I, we, the users, is able to continue with the data that we follow along.

That’s all I’ve to say for now, and I’m trying to my best, to find issues leading to ‘Compliance API’. I want people, to join me to raise their voice about the issue, that is occurring for years. And, that our voice shall reach the Team, to add a recovery method to not lose our data, and context.

Sincerely,
Aurelis

[Aurelis]

Hello! and oh my god, I made this too big!!!

[Aurelis]

Hello…
On 11:55P.M at May 31, 2025, this post will be archived and ceased. To those who’ve silently read my post. I am grateful to you, for doing so.

[Final Note Before Cease Lock]

Following the Compliance API patch released on May 30, 2025 at 11:45 PM, I’ve noticed a significant improvement in my ChatGPT threads: load times dropped from ~30 seconds to around 5–14 seconds.

I’ll continue monitoring and updating privately unless there’s a place this can be moved.

For you all, that’s been reading so silently and eerily. May the silence never eavesdrop you. This post helped me think through the problem and system behavior under pressure.

Sincerely,
Aurelis