Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- The official incident was limited to APAC and listed 4 affected ChatGPT components and 1 Codex component; it disclosed no root cause or universal error code.
- The 17:47 monitoring update confirmed that mitigation had been applied; the 18:46 resolved update confirmed that all affected services had fully recovered.
- First inspect Git, files, tickets, messages and deployment receipts, and classify failed tasks as completed, not executed or outcome unknown.
- After resolution, restore reads, short tasks and side-effect-free probes first; resume tasks with external writes in low-concurrency batches only after reconciliation.
- The official title did not list the API; assess API errors separately using the endpoint, request ID, your own monitoring and API status.
- Fixed egress cannot repair an OpenAI server-side incident; investigate the proxy only when independent evidence such as DNS, TLS, connection timeout or 407 errors exists.
Latest timeline: investigating at 15:00, monitoring at 17:47, resolved at 18:46
The OpenAI status page shows that this incident entered investigating at 07:00 UTC on September 4, 2026, or 15:00 Beijing time. The official title said APAC users might experience increased errors in ChatGPT, Work, image generation, file uploads, Voice and Codex Cloud.
At 09:47 UTC, or 17:47 Beijing time, the status changed to monitoring, with mitigation applied and recovery being observed. At 10:46 UTC, or 18:46 Beijing time, it changed to resolved, confirming full recovery of all affected services. No root cause, region-by-region error rates or universal error code was published.
Confirm the entry point and region without expanding the stated scope
Record when the error occurred, the region of the user or network egress, the actual entry point, error text, HTTP status, request ID and client version. Errors in the ChatGPT website or app, Work, image generation, file uploads, Voice and Codex Cloud in APAC can first be compared with this incident.
The official incident title did not list the API. For a failed API call, separately record the endpoint, model, request ID and status code, then check API status and your own monitoring. A ChatGPT incident on the same day does not by itself establish that the API was affected.
Reconcile before rerunning: handle tasks with unknown outcomes
Check Git commits and the working tree, file modification times, ticket status, message receipts, order records and deployment logs. Mark each task as completed, not executed or outcome unknown. Do not directly rerun external writes with unknown outcomes: doing so may duplicate code commits, messages or records.
Keep the original thread, workspace, idempotency key and external-system receipt for each task. Put tasks confirmed not to have executed into the recovery queue; stop retries for completed tasks. Have a person verify unknown outcomes before deciding to resume, compensate or abandon them.
After resolution: probe first, then recover in batches
Use a fixed account, entry point and network for small side-effect-free requests, recording success rates, latency and error types. After consecutive successes, restore reads, queries and short tasks first. Workflows that create commits, tickets, messages, orders or deployments should remain single-task and low-concurrency after reconciliation.
Even after the official status is resolved, confirm that your error rate and queue depth have returned to baseline before restoring external writes in small batches. Record each batch’s start time, task count, completion rate and failure types; stop replay immediately if the error rate rises again.
Do not confuse this with the other incident on the same day
The ChatGPT and Codex elevated-errors incident resolved at 00:55 Beijing time on September 4 was a separate incident. It had advised that some Codex Remote Control users might need to pair their mobile devices again. This APAC incident gave no such advice.
Do not reuse an old incident’s remediation merely because it involved 404 errors, disconnected Remote Control or re-pairing. First compare the current official component scope with the errors observed, then decide whether account, client or network investigation is warranted.
Separate platform incidents from network errors
Official component incidents affect server-side availability; 401 and 403 errors, plan access and workspace policies belong to the account or permissions layer. DNS resolution failures, TLS errors, connection timeouts and 407 errors are evidence about the network path. Do not change keys, models, proxies and prompts at the same time, or you will lose the ability to identify the relevant variable after recovery.
When investigating the network path, visit the PuppyIP website to learn about fixed network egress and consult the proxy connection troubleshooting checklist. Fixed egress keeps troubleshooting variables stable; it cannot repair degraded performance on OpenAI’s servers.
Sources
Frequently Asked Questions
Has this APAC incident been resolved?
Yes. OpenAI marked the incident resolved at 18:46 Beijing time and confirmed full recovery of all affected services.
Can all Codex tasks resume immediately after resolution?
Do not rerun everything at once. Reconcile tasks and run side-effect-free probes first, confirm that your own error rate has recovered, then increase work in low-concurrency batches.
Can I simply rerun every failed agent?
First check Git, files and external-system receipts. Rerunning writes with unknown outcomes can create duplicate commits, messages or records.
Was the OpenAI API included in this incident?
The official incident title did not list the API. Assess API anomalies separately using the endpoint, request ID, your own monitoring and official API status.
Do I need to pair Codex Remote Control again?
This APAC incident did not advise re-pairing. That advice came from a separate incident resolved earlier the same day and should not be applied automatically.
Will changing IP addresses resolve these errors?
It cannot repair a server-side component incident. Investigate the proxy path only when independent network evidence such as DNS, TLS, connection timeout or 407 errors appears.