ChatGPT Error Messages Explained: Causes and Fixes (2026)
Most ChatGPT error messages mean one of three things: an incident on OpenAI's side, a network or browser setting blocking the connection, or a usage limit. Check status.openai.com first, then refresh or start a new chat, then turn off VPNs, secure DNS and browser extensions and try another network. "Some tools are temporarily unavailable" is different: it is a tools outage, mostly in ChatGPT Work, that only OpenAI can fix (OpenAI Help Center and status page, 2026).
This guide explains every ChatGPT error message people search for in 2026, what each one actually means, and the fix that matches it. OpenAI documents only some of them, in its troubleshooting article updated September 16, 2026; for the rest, including "Some tools are temporarily unavailable," "Too many concurrent requests" and "Failed to read configuration layers," we traced the cause to OpenAI's status page, replies from OpenAI staff in the Developer Community, and bug reports, and we say which is which. The last sections cover keeping your work safe when ChatGPT fails, including with AI Toolbox (formerly ChatGPT Toolbox), a Chrome extension with 40,000+ users, a 4.7/5 Chrome Web Store rating, and a free plan.
By Adi Leviim, Co-Founder at Infi Developments. 7+ years building AI productivity tools. Last reviewed October 2026.
ChatGPT Error Messages: Quick Reference
Find your message in the table, then jump to its section. "Official" means OpenAI's Help Center explains it; "reported" means the explanation comes from OpenAI's status page, OpenAI staff replies, or user reports.
Broken config file in the ChatGPT desktop app or Codex
Fix or rename ~/.codex/config.toml
Reported
Download failed / File not found
A generated file expired or exceeded 512 MB
Regenerate the file
Official
"Some Tools Are Temporarily Unavailable" in ChatGPT
"Some tools are temporarily unavailable" means ChatGPT's tool layer, the part that creates files, reads uploads and runs code, is failing, while plain text chat still works. OpenAI's Help Center has no article for this message. The clearest record is the September 14, 2026 incident "Elevated errors affecting Work Mode in ChatGPT," in which OpenAI said "some ChatGPT Plus users using Work Mode experienced errors starting or resuming tasks, or had limited access to workspace tools and files," before applying a mitigation (OpenAI Status).
Users in the OpenAI Developer Community reported that "every user turn in a Work thread" showed the warning on the web and in the macOS app, and that the message "does not identify which tools are unavailable" (OpenAI Developer Community, September 14, 2026). What stopped working in that thread:
Creating downloadable files
Reading uploaded PNG and video files, and extracting ZIP files
Editing Excel files
Generating images from a reference image
How to handle it:
Check status.openai.com, especially the ChatGPT Work component. If an incident is open, nothing on your side will fix it.
Start a new thread once the status page shows recovery, and resend the request that failed.
Keep working in plain chat if you only need text. Text answers kept working during the incident.
Do not spend long on local fixes. In the community thread, a user who spent half an hour clearing the browser cache and moving to a new Work thread still saw the error until OpenAI's fix landed.
Custom GPT actions and apps can fail in similar ways with different wording, such as a GPT replying that its "API tools" are unavailable. For a custom MCP app that starts erroring after a tool changes, OpenAI notes the workspace "may still be using the previously approved tool list" (OpenAI Help Center).
"Something Went Wrong" and "Hmm...Something Seems to Have Gone Wrong"
"Something went wrong" is ChatGPT's general error: OpenAI says it "may indicate a temporary server issue or a problem with your local setup." OpenAI's fixes, in order (OpenAI Help Center, updated September 16, 2026):
Refresh the page or start a new chat.
Check status.openai.com for known outages.
Clear your browser cache and cookies.
Use an incognito or private browsing window.
Disable browser extensions, especially privacy or security tools.
Turn off VPNs or secure DNS tools like Web Protect.
Switch to a different network if possible.
"Hmm...something seems to have gone wrong." has no Help Center entry. In community reports it appears during OpenAI outages, such as June 10, 2025, and in 2026 alongside server errors when retrying an answer, so treat it like "Something went wrong" and check the status page first. On the iPhone and Android apps, "Something went wrong. Please make sure your device's date and time are set properly" is a device check error: OpenAI says it means "we were unable to generate a token for your specific device," and the fix is to correct the date and time, update the app and system, and check your connection.
"A Network Error Occurred" in ChatGPT
OpenAI says a network error, or "An error occurred while connecting to the websocket," means "your device is unable to establish a reliable connection to our servers." The full message users see reads "A network error occurred. Please check your connection and try again. If this issue persists please contact us through our help center at help.openai.com."
Turn off VPNs and proxies, and security filters like Web Protect.
Try another browser or a private window.
Switch network or device, for example from Wi-Fi to mobile data.
On a work network, ask IT to allow the ChatGPT domains and WebSocket traffic over TCP 443 to ws.chatgpt.com. OpenAI warns that if WebSocket traffic is blocked, "some features may stall, disconnect, or fail to stream updates correctly" (OpenAI Help Center).
It is not always your connection. A community report from September 2026 traced one network error to the server answering "Resume stream unavailable" after a long response broke, which is the same family of failure as "Error in message stream". If every fix fails and the status page is green, try a shorter prompt or a new chat.
Errors come and go; your chats should not. AI Toolbox (formerly ChatGPT Toolbox) keeps a searchable index of your ChatGPT conversations in your browser and exports any chat as TXT for free, or Markdown, JSON and PDF on Premium, so an outage never strands your work. 40,000+ users, 4.7/5 on the Chrome Web Store.
"Too Many Requests" and "Too Many Concurrent Requests"
Both are rate-limit messages, and in 2026 they usually appear during OpenAI incidents rather than because you used ChatGPT too much. Text chat has been unlimited on every plan since August 6, 2026, so neither message means you ran out of messages.
The 2026 wording: "Too many requests. You're making requests too quickly. We've temporarily limited access to your conversations to protect your data. Please wait a few minutes before trying again." A Plus user reported it on October 3, 2026 without sending any prompt. An OpenAI support reply on October 2 said "these errors may be linked to a recent outage."
"Too many concurrent requests" has no Help Center entry. Users have seen it during outages (January 2025, June 2025) and while ChatGPT was stuck on Thinking. OpenAI's agent article confirms limits "on the number of concurrent tasks" exist for ChatGPT agent.
What to do: close duplicate ChatGPT tabs, stop any running agent or Work tasks you do not need, wait a few minutes, and check the status page. If it persists for days, contact OpenAI Support through the Help Center with a HAR file.
"Too many requests in 1 hour. Try again later." is a 2022 to 2023 message from before paid plans had higher limits; we found no reports of it in 2025 or 2026. Tool limits are separate: uploads, images and voice have their own allowances, covered in the ChatGPT limits guide and the ChatGPT file upload limits guide.
"There Was an Error Generating a Response"
This is a single failed answer, and OpenAI's first fix is the simplest: "May be a one-time error, click 'Regenerate' button." If Regenerate fails too, OpenAI suggests restarting the browser or device, turning off VPNs, proxies and Web Protect, and trying a different browser, network or device.
If ChatGPT is stuck on Thinking or Generating instead, OpenAI's sequence is: wait 30 to 60 seconds, click Stop generating and then Regenerate, start a new chat and resend the prompt, hard refresh with Cmd or Ctrl+Shift+R, sign out and back in, then try a private window, another browser and another network. Long conversations are a common cause; OpenAI's own advice is to "start a new chat if the conversation is long or has many turns." The ChatGPT freezing guide explains why long chats slow the page down.
"You Don't Have Access to This Conversation"
The full message reads "You don't have access to this conversation. Make sure you're logged in to the right account, or ask the conversation owner to send you a share link." It usually means the chat belongs to another account or workspace, but in 2026 it has also appeared on people's own chats because of sync bugs.
Workspace links are private: a link created in a Business, Enterprise, Edu or Healthcare workspace "can be opened only by eligible members of the same workspace" (OpenAI Help Center).
Deleted chats break links: "Deleting a conversation removes its shared conversation link."
Desktop app mismatches: an OpenAI support reply on October 6, 2026 advised to "search for the chat on the web in the same workspace, and check Settings > Data controls > Archived chats," adding that "the access error alone doesn't confirm deletion."
Branches and shared projects: users reported the same permission error after branching a chat (May 2026) and in shared projects (September 2026, later fixed).
"Unable to Load History" in the ChatGPT Sidebar
"Unable to load history" means the sidebar's list of recent chats failed to load, not that your chats are gone. OpenAI has no Help Center entry, but OpenAI support replies in late September and early October 2026 gave the workarounds: "search, pinned chats and bookmarked direct links can be used as temporary workarounds," try another browser or network, disconnect a VPN or proxy, and try a platform other than the web.
Click Retry in the sidebar first.
Open Search (Ctrl+K or Cmd+K) and type a word from the chat; search loads separately from the list.
Open a chat from a bookmarked link or a pinned chat.
Try the desktop or mobile app while the web sidebar is failing.
This is where a second index helps. AI Toolbox's search runs on an index stored in your browser, so it can still find chats it has already indexed and open them by their direct link while the sidebar list is failing.
"Unusual Activity Has Been Detected From Your Device"
OpenAI shows this "when our systems detect traffic that may resemble automated or unusual behavior." The usual trigger is a VPN or proxy whose IP address is shared with many other users.
Disable VPNs or proxies "that might route your traffic through flagged IPs."
Restart your browser or device, or use a private window.
Make sure you are not sharing your account; change your password and secure your email.
If it continues and you use no bots or automation, contact OpenAI through the Help Center.
A related alert, "suspicious activity," is triggered by unusual sign-ins, inconsistent usage or "more simultaneous logins than usual." OpenAI's steps there add turning on two-factor authentication and logging out of all devices (OpenAI Help Center).
"Failed to Read Configuration Layers" in the ChatGPT Desktop App
This error comes from Codex, which became part of the new ChatGPT desktop app in 2026, and it means the app could not read its config file. It does not appear on chatgpt.com. Codex reads settings from several places, with personal defaults in ~/.codex/config.toml and optional project files in .codex/config.toml, and "the CLI and IDE extension share the same configuration layers" (OpenAI Codex docs).
Causes reported in OpenAI's Codex GitHub repository, with the fix for each:
What the error says
Cause
Fix
"unclosed table" or "key with no value" with a line and column
A typo in config.toml, or a file filled with blank bytes after a crash or forced restart
Open the file at that line and fix it, or restore a backup
Blank window, duplicate key
The same setting defined twice
Remove the duplicate line
"Permission denied (os error 13)"
Running codex with sudo made ~/.codex owned by root
Give the ~/.codex folder back to your user account
Back up ~/.codex/config.toml before you edit it. As a last resort, renaming the file lets the app start with default settings, but you lose your custom settings, as one user who deleted it warned. If you only use chat, ChatGPT Classic, the previous desktop app, keeps working while you fix the new one.
Legacy Messages You May Still Read About
"ChatGPT is at capacity right now": the 2022 to 2023 overload page. We found no reports of it in 2025 or 2026. The 2026 equivalent is "Selected model is at capacity. Please try a different model," for which an OpenAI support reply suggested a fresh chat with the same model.
"Too many requests in 1 hour": replaced by the newer "Too many requests" wording above.
"Download failed" or "File not found" on a generated file: OpenAI says generated files "expire quickly" and must be under 512 MB, so ask ChatGPT to create the file again.
Native ChatGPT vs AI Toolbox When ChatGPT Fails
No extension can fix an OpenAI outage; what you can control is whether your work survives one.
Task
ChatGPT (built in)
AI Toolbox Free
AI Toolbox Premium
Find a chat when the sidebar fails
Search, pinned chats, direct links
Search on a local index, first 5 results
All results plus Advanced Search filters
Keep a copy of an answer
Copy by hand; full account export by email
Export any chat as TXT
TXT, Markdown, JSON, PDF, in bulk
Spot a chat that is getting too long
Not shown
Context Window Meter
Meter plus handoff to a fresh chat
Price
Included
Free
$9.99/month or $99 lifetime
AI Toolbox is also a browser extension, so when you troubleshoot, turning extensions off is still the right test. If ChatGPT works with AI Toolbox off and fails with it on, tell us so we can fix it. The ChatGPT export guide and the backup guide cover saving chats before you need them.
Limitations
OpenAI documents only five of the messages above in its Help Center. For the others, the explanations come from OpenAI's status page, OpenAI support replies in the Developer Community and user bug reports, which describe specific incidents and may not cover every cause. Exact wording also changes between the web, desktop and mobile apps. When a message persists across browsers, networks and devices while the status page is green, OpenAI Support is the only route, and it asks for a HAR file recorded while the error happens.
Frequently Asked Questions
What does "Some tools are temporarily unavailable" mean in ChatGPT?
It means ChatGPT's tools, such as file creation, reading uploads, spreadsheet editing or image generation, are failing while text chat still works. It appeared on every turn in ChatGPT Work threads during OpenAI's September 14, 2026 incident. Check status.openai.com, wait for a fix, then start a new thread.
Why does ChatGPT keep saying something went wrong?
"Something went wrong" is a general error that OpenAI says can be a temporary server issue or a problem with your local setup. Refresh or start a new chat, check status.openai.com, clear cache and cookies, try a private window, and turn off extensions, VPNs and secure DNS tools.
How do I fix a network error on ChatGPT?
OpenAI says a network error means your device cannot keep a reliable connection. Turn off VPNs, proxies and security filters such as Web Protect, try another browser or a private window, and switch to a different network or device. On work networks, WebSocket traffic to ws.chatgpt.com must be allowed.
What does "Too many concurrent requests" mean on ChatGPT?
It is a rate-limit message that OpenAI does not document. Users mostly report it during outages or when ChatGPT is stuck on Thinking. Close duplicate ChatGPT tabs, stop tasks you do not need, wait a few minutes, and check status.openai.com. It does not mean you ran out of messages.
What does "Failed to read configuration layers" mean?
It is an error from Codex in the new ChatGPT desktop app, CLI or IDE extension, meaning the config file could not be read. Common causes are a typo in ~/.codex/config.toml, a duplicate setting, a file corrupted by a crash, or a folder owned by root after running with sudo.
Why does ChatGPT say I don't have access to this conversation?
You are usually signed in to a different account or workspace than the one that owns the chat, or the chat or its link was deleted. Workspace links open only for members of that workspace. Search for the chat on the web and check Archived chats; OpenAI says the error alone does not confirm deletion.
What does "Unable to load history" mean in ChatGPT?
The sidebar's list of recent chats failed to load; your chats are not deleted. Click Retry, use Search, open a chat from a bookmarked link, or try the desktop or mobile app. OpenAI support has called search, pinned chats and direct links temporary workarounds while it fixes the list.
Bottom Line
Read the message, then match the fix: an incident means wait, a network error means change the connection, a single failed answer means Regenerate, and a desktop config error means fixing one file. "Some tools are temporarily unavailable" and the rate-limit messages are mostly OpenAI's to fix. Keep copies of the chats you rely on so the next outage costs you time, not work.
A Full Stack Developer with 7+ years of experience building AI productivity tools. Leads product development and frontend architecture for AI Toolbox, the Chrome extension suite (ChatGPT, Gemini, Grok, and Claude modules) that helps users search, organize, and export their AI conversations.
Liked this guide? Get the next one first.
Join 40,000+ readers getting practical prompts, new features, and subscriber-only deals. No spam, unsubscribe in one click.
New features and product updates before anyone else
Time-saving prompts and workflows for ChatGPT, Gemini, Claude, and Grok
Adi Leviim