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 resolution was October 12, 2026 at 16:02 UTC, or October 13 at 00:02 in Asia/Shanghai.
- The affected channel published Costs data through the Compliance API; this was not an outage of all OpenAI services or billing.
- Service recovery and complete downstream reporting are separate checks. Verify gaps, pagination and sync boundaries before deciding on backfill or support.
- The public incident promises no refunds, compensation or usage resets, and provides no per-workspace reconciliation results.
Which reporting channel has recovered?
OpenAI's status page now marks Delayed Costs data in the Compliance API as Resolved and says the affected services have fully recovered. The exact resolution was October 12, 2026 at 16:02:47 UTC, equivalent to October 13 at 00:02:47 in Asia/Shanghai. Recovery of the reporting channel does not establish that a particular invoice has been reconciled.
The first report appeared on October 9 at 17:23 UTC, initially using the broader phrase Costs API. The 17:27 UTC clarification narrowed the issue to Compliance API Costs data and explicitly excluded the API Platform organization costs endpoint and ChatGPT usage. Later updates also said API Platform requests were unaffected.
If your problem is failed model requests, unavailable ChatGPT or a subscription limit that has not reset, investigate that product and error separately. This reporting incident does not establish the cause.
The Sunday backlog forecast is now historical
The October 9 Monitoring update at 22:03 UTC said mitigation was processing delayed data, some customers could still see incomplete results, and catch-up was expected through Sunday, October 11. That described the expectation at the time; it is not evidence that a backlog remains today.
The October 12 resolution is the newer official status. Keep the first report, scope clarification, mitigation and resolution in the incident timeline. Do not present the earlier forecast as a current deadline or turn an aggregate recovery statement into proof that every workspace has received every record.
Separate compliance reporting from organization costs
The Compliance Platform serves ChatGPT Enterprise and Edu workspaces. Access needs a workspace-scoped Admin key with appropriate permissions; personal subscriptions do not automatically qualify.
The ordinary organization costs endpoint is a separate channel. Its public API reference lists GET /organization/costs, with /v1/organization/costs in the request example. OpenAI explicitly excluded it from this incident. Shared use of the name Costs does not make their identities, permissions or reporting results interchangeable.
For example, a company might sync a compliance report from the Compliance API while checking API spending through the ordinary organization endpoint. A gap in the first report does not show that the second channel failed or that actual spending was zero. This is an illustrative scenario, not customer data we accessed.
Check four things in your own integration
First, identify the data source. Use existing integration documentation and authorized reports to record the workspace, data category, retrieval time and timezone. Establish that you use the affected channel before matching a gap to the incident. The official report supplies no country-by-country impact list.
Second, locate the incomplete period. Compare existing sync checkpoints, job logs and reports using the same definitions. Distinguish records that have not arrived, failed reads and queries that genuinely contain no records. One successful request or a plausible report total does not prove continuous coverage.
Third, check time boundaries and pagination against that interface's documentation. For the ordinary organization costs endpoint, start_time is inclusive, end_time is exclusive, and, if another page is available, its next_page cursor is passed as page to continue. These fields belong to the ordinary endpoint; they are not Compliance API parameters or an incident-specific fix.
Fourth, backfill only a confirmed gap through your integration's established sync and deduplication process. Preserve the before-and-after ranges and results. We have not initiated a backfill; the integration owner should handle retries, permissions and data processing through their existing procedure.
If gaps remain, contact the actual integration provider
Contact your integration partner first, with the workspace owner involved. Provide the workspace ID, error, and start date, time and timezone through the existing support channel.
The resolution supplies no per-workspace completeness checklist, compensation, refund promise or usage-reset arrangement. It announces no new fee or regional eligibility change. The official incident is resolved; whether your data is complete still depends on your integration's actual checks.
Sources
Frequently Asked Questions
Why use Compliance API when the original report said Costs API?
The same incident clarified its scope about four minutes after the first report. Use the newer precise description while preserving the timeline. The initial broader wording does not establish that the ordinary organization costs endpoint was affected.
Can this incident explain zero API costs in my report?
The name alone cannot establish that. Check the source, permissions, time window, pagination and actual records. OpenAI explicitly excluded the ordinary organization costs endpoint, so a zero value requires its own evidence.
Does Resolved mean I need to generate a new Admin key?
The incident requires no key rotation or broader access. Check permission errors through existing roles and integration guidance.