Hi all — our published Custom GPT with OAuth began hanging on *“Talking to ”*
I can see that there was a global outage, it has been marked as resolved
But we can still see the issue persist
I have tried interacting with other popular GPTs like Canva, i’m seeing the same issue
Is there any issue, when will it be solved ?
Outbound JSON POSTs from GPT Actions No Longer Working?
Hi Fabulous Dev team
Looking for an update on this. My GPT in development started hanging yesterday — it just shows “Talking to {domain}” and never completes — after previously working fine with no changes on my server side. The domain is validated.
Can a dev confirm whether outbound JSON POSTs from GPT Actions have been blocked? If so, what is the recommended path forward to overcome this limitation?
My current flow is straightforward:
Collect a few data points from the user
Submit them to my server via JSON POST to create a job
I went this route because of the GPT marketplace opportunity, but if a GPT can no longer send metadata to an API endpoint, that effectively destroys my business plan.
I’ve already adapted once — first using base64 blobs (later restricted), then switching to metadata-only POSTs (which now appear to be disabled). Each rollback makes it harder to justify further investment. I’m now researching the Assistants API, but if no one can send metadata via GPT connectors anymore, then the value proposition of this feature feels compromised.
Right now, my confidence in the platform’s ability to support developers and marketplace growth is limited. Many opportunities to implement your products are being removed, which is concerning for anyone building a business on this ecosystem.
Thanks 
Same things happening since yesterday
I was able to reproduce the issue of an action flow being stuck in the ‘talking to’ stage and will raise this with OpenAI.
Same here. My custom GPT is stuck on “talking to domain”. And just confirmed that it’s not only my custom GPT, but public GPTs like Wolfram or Expedia have a similar symptom (the talk with domain finishes immediately and no response from the action)
Same here - all Actions of ALL GPTs are stuck indefinitely on ‘Talking to …’. This issue occurs with any GPT, regardless of the chosen model.
Can someone from OpenAI look into this? This is such a critical issue that it affects all GPTs using Actions.
According to the stats I see, the issue started at 10 p.m. UTC and is still ongoing.
Attached example screenshots of the issue:
Lot’s of things are broken, when using Custom GPT.
In this case, go to privacy settings when in the Custom GPT and switch to ‘always allow‘.
Try it hard, because even the default <select> element seems not working all the times.
This will raise → "is_consequential": false
In the debug window you can see it’s waiting for confirmation, which is skipped when you choose always allow by default (at least it’s working for me).
In the mobile app, request for connection is never ever asked, you there it must be set to always allow since ages.
But my Custom GPT’s are more broken then this. My server sends the data requested, the GPT fetches it (I’ve doublechecked all) and then it’s not using it. It is hallucinating as hell (All 5’s and 4’ on both web and app).
Also the @foobar mentions within a Custom GPT are not working since a month. It’s all so broken.
Glad to see that I’,m not the only one. I thought this would get corrected after the update last night. Even though openAI claims no issues are impacting the service, somehow it seems thiat this get overlooked.
If anybody has had it manage to finally self correct please let us know. This is reallly annoying and it doesn’t seem like it’s even flagged as an issue. Anybody got an official response fro openAI?
I’m getting the same thing. Coincidentally, it was while I was doing an update and proceeded to spend hours trying to resolve an issue that wasn’t on my end. Those are 4-5 hours I won’t get back. God knows how much more I messed things up working on a non-issue. Hopefully it gets resolved soon.
Just to be clear. I get the “Talking to {domain}” hang. Oddly enough if you ask it, “did you find an answer yet”, you will get the answer. So the data pulls, it just doesn’t automatically respond.
Setting an action to Always allow does not resolve issues with Actions that require OAuth (i.e., GPTs that have a Connected accounts tab in Privacy settings). Actions with OAuth remain stuck indefinitely on ‘Talking to…’.
Example screenshots:
I’ve been working on it for 16 hours straight.
The custom action connects to my server without problems ("is_consequential": false).
And it does receive data from my server. But then the GPT comes with a totally wrong answer.
Like server is saying “apple juice is good for you” and GPT says it received “32 is a nice integer and I do hear the music loud“.
It’s so broken since about 24 hours.
Updated: Response from openAI support
Thank you for reaching out to OpenAI support and for sharing the details of your experience. I understand how disruptive and frustrating this issue has been, especially given the widespread reports from multiple developers and the impact on business use cases.
We’ve noted that Custom GPTs using Actions — particularly those requiring OAuth — are hanging indefinitely at “Talking to …”, with POST/PUT/DELETE requests failing, while GET requests sometimes succeed under specific settings. We also understand that workarounds such as “Always allow” permissions have proven ineffective, and that hallucinations or missing server data may occur.
While there is currently an active outage listed on our status page, it does not appear to be related to the issue you’re encountering. However, we recognize this may be part of a larger problem that is not yet reflected there. To help us investigate further and escalate appropriately, could you please provide the following details:
A description of the exact issue you are experiencing
Steps to reproduce the problem
Any screenshots of the error or affected interface
Once we receive this information, we’ll be able to review your case for deeper investigation.
I’ve sent the support rep the link to this thread so at least we are now on their radar.
Same issue here — really relieved to find this thread.
I’ve been debugging since this morning. Checked my domain, DNS, and all my endpoints. On the server side, everything works as expected.
I tested a few features and here’s what I found (hopefully this helps the team investigating this issue):
The key problem seems also be a UI issue. Previously, whenever a request involved login or modifying/deleting data, a “Login to your website” or “Confirm” button would appear. That button is now gone, which means users can no longer authorize these requests. As a result, every request that requires modification or deletion now returns something like this:
{
"function_name": "delete_project_api_projects__project_id__delete",
"domain": "my-domain-name",
"message": "The requested action requires approval",
"action_id": "g-26a1aa9cd5f6d11cfabdaa173e5abf572cd4ef21"
}
Since the button has disappeared, users cannot log in or grant the necessary permissions to proceed with the Action.
As a workaround, I found that going through GPT → See Details → Privacy & Safety → Connected Accounts → Login allows login to succeed.
Up until about an hour ago, I was still able to run GET requests. But after re-logging in, even GET requests no longer work.
Update:
Now even GET requests are returning the same approval message, despite being marked as non-consequential:
// [debug] Calling HTTP endpoint
{
"domain": "matt.youseminar.cn",
"method": "get",
"path": "/api/projects/",
"operation": "get_my_projects_api_projects__get",
"operation_hash": "d4e6bf0979b7677c3f8618f4dd162ad539494f54",
"is_consequential": false,
"params": {}
}
// [debug] Response received
{
"function_name": "get_my_projects_api_projects__get",
"domain": "matt.youseminar.cn",
"message": "The requested action requires approval",
"action_id": "g-26a1aa9cd5f6d11cfabdaa173e5abf572cd4ef21"
}
This makes it impossible to perform even basic read operations. It seems like the approval flow is being incorrectly applied globally, not just to destructive actions.
Update Again
Thanks to @dkakeith for pointing out @iyvljtjqleyj’s post. I clearly remember setting the Privacy Settings to always allow when I first started using this GPT. But after double-checking just now, I noticed that on the web version it had somehow reverted back to ask. I switched it back to always allow again.
Now the GET requests are working for me — but PUT, DELETE, and POST are still not functioning.
This custom GPT is something my friends and I use every day, so it’s quite concerning. Right now I don’t even know where the issue lies or when it might be fixed.
Experiencing the same issue. My GPT Action is permanently stuck on the “Talking to“ stage.
Had the same issue for the past 2 days and was tearing my hair out until I saw @iyvljtjqleyj’s post above about changing the Privacy Settings. Worked for me! In case you’re a newbie like I am, here’s a screenshot to show you where to find the settings:
Note: the workaround above seems to help only with GET requests. POST requests will still hang indefinitely without asking for permission as of 60 seconds ago.
I picked one of the top recommended GPT that would use an action and tried - also stuck at “Talking to chat.consensus.app”, going on 10+ minutes
Seems actions are just broken.
Confirmed this is not just an OAuth issue: made routes on my api that were zero auth, bearer token, and Oauth. for all 3, GET will work OK given the “always allow” workaround. POST/PUT/PATCH/DELETE all hang no matter what.
I’m experiencing the same issue.
GET requests work fine, but POST requests get stuck on “Talking to ”.
Everything works correctly when tested outside GPT (e.g. with Postman).
Please let us know if there’s any update.
Happy to share more details if helpful.
Thanks in advance!
Can confirm having the same issue for over 24 hours now. My custom gpt is for personal use and it’s frustrating enough - I can’t even imagine using it for business use if the radio silence from open AI is the norm on such things. Wild that this is the leading company in the field…
