Image generation, agents, access & developer workflows · Daily incidents

AI incidents: OpenAI image errors, Workspace Agent recovery and GitHub disruption — 12 November 2026

OpenAI resolved its image-model, Workspace Agent and login issues. GitHub, Fireworks and Bolt also report resolved disruptions.

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.

OpenAI image recovery confirmed after the original cutoff

13 November follow-up: OpenAI posted resolution at 12:25 Istanbul on 12 November, five minutes after this edition’s cutoff.

OpenAI applied a mitigation for elevated errors on GPT-image-2.5 and 2.6 Sunburst and Flare and was monitoring recovery at this edition’s cutoff. It separately resolved problems starting Workspace Agent runs and a short login, ChatGPT-feature and API-conversation incident. GitHub, Fireworks and Bolt also published resolved notices.

This scan covers 9 November at 11:53 through 12 November at 12:20 Europe/Istanbul (UTC+3), including the 48-hour overlap from the previous successful scan. Previously covered events are deduplicated. These are separate operator notices, not evidence of one shared cause. Resolution does not prove every failed generation or task was subsequently completed.

Affected service Status at cutoff Notice milestones, Istanbul
OpenAI GPT-image-2.5 / 2.6 Sunburst and Flare Monitoring 12 Nov: acknowledged 11:52; scope expanded 11:58; mitigation/monitoring 12:18
OpenAI Workspace Agent starts Resolved 12 Nov: acknowledged 01:50; resolved 11:32
OpenAI logins, selected ChatGPT features and API conversation creation Resolved 11 Nov: acknowledged 23:43; resolved 23:52
GitHub Actions, Pages, Copilot and Webhooks Resolved 11 Nov 23:48 acknowledgment → 12 Nov 00:07 monitoring → 00:21 resolution
Fireworks GLM 5.3, Kimi K3.1 Pro and GLM 5.3 Fast Three resolved notices Separate component intervals detailed below
Bolt Prompting Four resolved notices Separate intervals on 11 and 12 Nov; not one continuous outage

Times above are notice milestones. Actual earliest user impact is unknown unless explicitly stated. Durations below identify exactly what the provider record measures.

OpenAI image models: mitigation applied, recovery still being monitored

OpenAI reported elevated errors on its API Images component for GPT-image-2.5 Sunburst and Flare at 11:52 on 12 November. At 11:58, it expanded the named scope to include GPT-image-2.6 Sunburst and Flare. At 12:18, OpenAI said it had applied a mitigation and was monitoring the affected services to confirm recovery. The notice had not reached resolved status at our 12:20 cutoff. Original OpenAI image incident.

The notice does not establish an outage of all ChatGPT features or every image model. It supplies no technical root cause, regional breakdown, source-supported workaround or independently established impact start. We therefore do not label the interval since acknowledgment as total user downtime.

Workspace Agent starts recovered after an overnight disruption

OpenAI says users can start Workspace Agent runs again. The incident opened at 01:50 on 12 November and resolved at 11:32. Its affected component is ChatGPT Agent; the described symptom is inability to start new Workspace Agent runs, not a verified failure of every running task. Workspace Agent notice.

The provider-recorded component interval is 01:50:17–11:32:37 Istanbul, approximately 9 hours 42 minutes. That measures the interval in OpenAI’s component record; the first affected user request could have occurred earlier. OpenAI did not publish a cause or workaround in the retrieved notice.

A separate OpenAI login and conversation issue resolved on 11 November

OpenAI resolved a separate disruption involving logins and selected ChatGPT features at 23:52 on 11 November. The first acknowledgment was at 23:43. Updates described logins, signups, Work, dots, sites and spaces; API conversation-creation errors were added to the notice at 23:50. Login and conversation incident.

Component markers were added after the first notice, so their shorter interval must not be substituted for actual incident duration. Earliest impact, cause and regional scope remain unknown. This is distinct from the longer Workspace Agent event and today’s image-model errors.

GitHub: Copilot and developer workflows recovered across midnight

GitHub resolved a multi-product incident affecting Actions, Pages, Copilot and Webhooks. The first notice was published at 23:48 on 11 November. GitHub moved to monitoring at 00:07 on 12 November, then reported that the remaining backlog had been processed and resolved the incident at 00:21. GitHub incident.

The full timeline crosses midnight in Istanbul. Resetting it at midnight would obscure the sequence. The notice supports impact to the named GitHub products; it does not establish an outage of Microsoft’s separate consumer Copilot service or a failure at an underlying model provider. GitHub said a root-cause analysis would follow; none was present at cutoff. No user workaround was documented.

Fireworks: three hosted-model degradations resolved

Fireworks recorded separate disruptions for GLM 5.3, Kimi K3.1 Pro and GLM 5.3 Fast. These describe Fireworks’ hosting components, not outages across Z.ai’s or Moonshot’s own services.

Model Recorded interval, Istanbul Status / notice
GLM 5.3 11 Nov, 15:23:27–15:47:50 About 24 minutes of recorded degradation; resolved. Notice
Kimi K3.1 Pro 12 Nov, 00:40:27–00:43:19 About 3 minutes in the component record; resolved. Notice
GLM 5.3 Fast 12 Nov, 07:30:27–08:00:53 About 30 minutes of recorded degradation; resolved. Notice

A later 03:30 resolved update for Kimi K3.1 Pro describes elevated latency for some customers. That later publication time is not the component recovery boundary; it does not justify extending the recorded disruption until 03:30. None of these notices supplies a detailed cause or workaround.

The older deployment, training and batch-inference incident still retains its 6 November restoration/monitoring update. The separate older Kimi K3 notice remains unavailable in the retrieved detail/history, as explained in yesterday’s correction. No new recovery claim is inferred from that absence.

Bolt: four separate Prompting notices are resolved

Bolt recorded four separate degraded intervals for its Prompting component:

These are provider-recorded component intervals, not independently measured earliest user impact. The notices do not give a technical cause, region-specific impact or workaround. They should not be combined into one uninterrupted outage. A separate completed maintenance notice is not evidence that maintenance caused these events.

Sources

Original incident notices are linked beside each finding. The primary records include OpenAI’s image incident and GitHub’s multi-product incident. We rechecked Anthropic’s Claude history, OpenAI’s history, Google Cloud/Workspace and the rendered AI Studio history, and ElevenLabs’ regional histories. No additional fresh qualifying Anthropic, Google or ElevenLabs event was established beyond previously published coverage. That does not establish uninterrupted service for every user.

The wider creative-tool watchlist and accessible original customer reviews were also checked. Isolated complaints without matching independent evidence remain unpublished; inaccessible review pages and client-rendered release listings limit coverage. See our reporting method, previous incident report and separate weekly product-news roundup.

Update timeline