Payments, AI APIs, voice agents & coding workflows · Daily incidents
AI incidents: OpenAI payment declines, DeepSeek recovery and Fireworks restoration — 7 November 2026
OpenAI resolves U.S. business Mastercard declines; DeepSeek and Grok Bot recover. Fireworks restores services, and GitHub confirms Copilot agent delays.
Publisher disclosure: MUVO offers competing music, remix, and video tools. This report summarizes public sources and provider documentation; it is not an independent product test or a claim that MUVO is free of similar issues. Read our reporting method.
Incident summary and status at the cutoff
Update, 11 November: OpenAI has resolved the payment-decline incident affecting some U.S. business Mastercards. DeepSeek and Grok Bot have reported recoveries, while Fireworks says its older deployment, training and batch-inference disruption is restored and under monitoring. GitHub’s newly updated postmortem confirms earlier Copilot coding-agent delays. ElevenLabs has also documented two resolved EU voice-agent incidents.
This edition covers 4 November at 12:15 through 7 November at 14:35 Europe/Istanbul (UTC+3), including a 48-hour overlap from the last successful scan. Earlier incidents appear only as labeled follow-ups or recurrence context. All event times below are Istanbul time. Provider recovery claims do not establish that every account or previously failed task recovered.
OpenAI: U.S. business Mastercard payment incident resolved
OpenAI says some U.S. business Mastercards have suffered payment declines that can block API automatic recharges and one-time credit purchases. If a prepaid balance is exhausted, API calls may then fail. The notice identifies card-number prefixes 544015, 546400, 543502, 543799 and 512998, plus some cards starting 513989. It does not describe a general ChatGPT outage or failure of all Mastercard payments. Original OpenAI notice.
- 6 November, 21:38: First acknowledgment; the status label is identified, while the message describes investigation.
- 6 November, 21:50: Scope and payment advice reiterated.
- 7 November, 04:29: Mitigation applied; monitoring began to confirm recovery.
- 10 November, 19:06: OpenAI marked the incident resolved; added to this report on 11 November.
During the disruption, OpenAI recommended adding a card from another issuer before the next payment, and using that card to retry an already failed credit purchase. It said it was working with payment partners. The technical cause and earliest impact remain unknown. The resolution notice does not establish that every earlier failed purchase completed automatically. See the 11 November recovery roundup.
DeepSeek: chat, search and two API services recover on 7 November
DeepSeek marked its partial web/API interruption resolved at 10:56 on 7 November. The affected list names V4 Pro API, V4.1 Flash API, chat and search. Its first visible update at 10:11 already says a fix was implemented and monitoring had begun. DeepSeek incident, official Atom feed.
The page displays an approximately 45-minute interval from monitoring to resolution. That is not evidence of the full outage duration: an earlier impact start is not supplied. The notice gives no cause, affected plans, region-specific scope or user workaround. This is separate from the 3 November DeepSeek recovery.
Fireworks: older services restored; seven GLM endpoint incidents resolved
Deployment, training and batch inference: recovery update
Fireworks announced restoration at 21:28 on 6 November and moved its older service-degradation incident to monitoring. The first notice, at 23:51 on 30 October, concerned deployment creation, dedicated training and batch inference. The recovery message names dedicated deployments, training and batch inference. Continuing Fireworks notice.
The provider directs remaining problems to its account team or support. It supplies neither a root cause nor complete impact boundaries, so the elapsed time between notices must not be described as a continuous week-long outage. We added recovery notes to the 4 November, 5 November and 6 November editions.
Hosted GLM endpoints: separate component intervals
Seven additional Fireworks notices report impaired performance, followed by resolution. The intervals below come from its component-impact records, rounded to the displayed minute. They are separate incidents, not continuous downtime between rows, and concern Fireworks hosting rather than the model developer’s own service.
| Endpoint and original notice | Recorded interval, Istanbul |
|---|---|
| GLM 5.3 Fast | 6 Nov 15:31–15:44 |
| GLM 5.3 Flash | 7 Nov 02:04–02:10 |
| GLM 5.3 Flash | 7 Nov 03:11–03:21 |
| GLM 5.3 Flash | 7 Nov 03:33–05:40 |
| GLM 5.3 Flash US | 7 Nov 04:54–05:18 |
| GLM 5.3 Fast | 7 Nov 05:09–05:19 |
| GLM 5.3 Fast | 7 Nov 11:22–11:36 |
These notices do not disclose a technical cause or workaround. Exact timestamps and component identifiers are retained in the evidence ledger.
ElevenLabs EU: retrospective audio failures on 3–4 November
ElevenLabs reports that some EU Agent conversations using Eleven v4 Turbo produced choppy audio with missing speech. Both notices are resolved. Newly verified in today’s scan, they describe earlier events rather than a new 7 November outage.
- 3 November, 13:10–14:05: A 55-minute provider-recorded affected interval. First incident.
- 4 November, 12:15–14:15: A two-hour recurrence, beginning at this edition’s research-window boundary. Second incident.
ElevenLabs attributes the failures to the model producing audio slower than real time under heavy load. It reports improving throughput and adding earlier alerts. These are provider explanations, not MUVO measurements. The notices concern EU voice-agent conversations using the named model; they do not establish a failure of all music, speech or video products.
The pages carry event-time metadata, but we could not independently establish when each retrospective notice first became public. Our 6 November coverage note now points to this additional evidence.
Grok Bot: two distinct recoveries on 6 November
The official xAI feed records a computer-use problem and a later general Grok Bot degradation, both resolved. Original status feed.
- Computer use: Acknowledged at 14:49, resolved at 15:03. The provider specified some regions without naming them and said other Bot features remained normal. Computer-use notice.
- Grok Bot degradation: Acknowledged at 16:57, resolved at 18:15. No more specific symptom or regional scope was supplied. Bot notice.
These are notice milestones, not independently established impact boundaries. Neither event supplies a technical cause or workaround, and they should not be joined into one continuous Grok outage.
GitHub: postmortem confirms earlier Copilot coding-agent delays
GitHub’s summary, updated on 7 November, connects two 4 November incidents to an event-routing configuration change. It now explicitly includes delayed Copilot coding-agent sessions and Code Scanning analyses alongside Actions, Pages and Webhooks. Combined provider summary, linked recurrence.
GitHub records two impact periods on 4 November: 18:49–19:48 and 20:20–20:55—59 and 35 minutes respectively. It says incorrectly configured feature flags paused event delivery; the first rollback left code that a later deployment reintroduced. A second rollback and backlog processing restored service. GitHub says delayed events were processed and describes planned flag validation, safer fallback behavior and alerting improvements.
This is a postmortem update to 5 November’s coverage, not a new 7 November failure.
Bolt: custom-domain status reporting restored
Bolt resolved a problem reporting custom-domain status at 21:13 on 6 November. Investigation began at 18:26, with the specific symptom clarified at 18:36. Bolt incident.
The notice concerns status reporting. It does not establish that custom-domain websites were unreachable, that publishing failed, or that prompting was affected. Cause and workaround were not supplied.
Sources
Original notices are linked beside each event. We rechecked Claude’s incident history, Google AI Studio’s rendered history, Google Cloud history and ElevenLabs regional histories separately. No additional new Claude or Google event was established in this scan; that does not rule out unreported account-level failures.
The wider creative-tool scan included original reviews and community accounts. Some review sources blocked retrieval, and several status pages exposed incomplete archives. Isolated allegations were retained as unpublished leads; positive customer reports of restored access were also considered. See our reporting method. Product announcements remain separate, for the weekly news edition.
Practical analysis: Check the precise product and job or payment state before repeating a failed action that could create duplicate work or charges. A provider’s recovery announcement does not prove that an earlier task was replayed or that every account’s access has returned.
Update timeline
-
11 November 2026: Added OpenAI’s 10 November 19:06 Istanbul payment resolution; retained the original edition date and earlier timeline.
-
7 November 2026, 14:35 Istanbul / 11:35 UTC: Research cutoff. Verified payment monitoring, service recoveries, retrospective voice-agent incidents and GitHub’s postmortem update.
-
7 November 2026: Published this edition and added dated updates to the 4–6 November reports, preserving their original publication dates and cutoff accounts.