AI incidents: ElevenLabs sign-in issue, Runway and DeepSeek recoveries — 4 November 2026

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 publication

ElevenLabs is investigating password-based sign-in failures, while Runway has resolved delayed or failed Seedance generations and DeepSeek has restored its affected chat and API services. Bolt prompting remains degraded. Separate recoveries affect Copilot, Cursor cloud workflows, Descript avatars, Cerebras and Fireworks model endpoints.

The research window is 1 November at 12:02 through 4 November at 11:54 Europe/Istanbul (UTC+3). All times below use Istanbul time unless labeled otherwise. Events on 2–3 November retain their original dates; an older unresolved Fireworks notice is explicitly identified. These are provider reports, not MUVO tests of every account’s recovery.

ElevenLabs: password sign-ins are intermittently failing

ElevenLabs’ Global (United States) status page published an investigating notice at 11:21 on 4 November. It identifies password-based sign-ins and explicitly says other sign-in methods are working. No recovery notice was present when rechecked for this edition. Original incident.

The notice does not name an underlying cause, an impact-start time, affected plans or an API-generation failure. Its regional status-page context should not be expanded into an outage of all ElevenLabs services worldwide. An existing alternative sign-in method may remain usable, according to the provider; the notice does not promise that changing an account’s authentication settings will fix it.

Runway: Seedance video generations have recovered

Runway marked its Seedance generation incident resolved at 07:55 on 4 November. The first notice appeared at 02:49, followed by an investigation update at 02:55 and a fix/monitoring update at 03:21. The reported symptom was failed or delayed video generations. Runway incident and timeline.

Runway did not identify a Seedance version, region, technical cause or specific workaround. The notice concerns Runway’s integration; it does not establish a ByteDance-wide outage. The roughly 5 hours 6 minutes from first notice to resolution is a notice interval, not a measured duration for every failed generation. The earlier Runway Kling disruption is a separate event.

DeepSeek: chat and two API model services restored on 3 November

DeepSeek reported partial web/API unavailability and then restoration on 3 November. The affected components were Chat Service, V4 Pro API and V4.1 Flash API. Its timeline begins with investigation at 11:58, enters monitoring at 12:06, and ends with resolution at 13:48. DeepSeek’s original notice.

The provider’s dashboard labels the event 1 hour 49 minutes; its displayed minute-level endpoints are 08:58–10:48 UTC. We retain that provider label rather than calculate a different duration from rounded times. Repeated monitoring updates do not establish separate new outages. No technical cause, regional breakdown or workaround is disclosed.

GitHub and Cursor: Copilot errors and slow cloud-agent starts resolved

Copilot’s GPT-5.4 Nano errors on 2 November

GitHub reported degraded GPT-5.4 Nano availability in Copilot products and IDE surfaces. Its notice ran from 20:46 to 21:24 on 2 November, with mitigation/monitoring at 21:10. GitHub attributed the issue to an upstream model provider without identifying a technical cause. This does not establish that every OpenAI API model was unavailable. GitHub incident.

API, Git and Copilot disruption on 3 November

GitHub’s separate incident was acknowledged at 16:25 and resolved at 16:56 on 3 November. Updates also identified Issues degradation, Copilot mitigation at 16:51 and normal API operation at 16:52. The provider promised a later root-cause analysis. Official timeline.

Git operations delayed Cursor Cloud Agents and Bugbot

GitHub’s evening Git Operations incident was resolved at 01:41 on 4 November, following a first notice at 23:57 on 3 November and monitoring at 01:37. GitHub notice.

The Cursor-related Bugbot and Cloud Agents status notice separately ran from 00:40 to 01:53 on 4 November. It attributed slower agent startup and Bugbot runs to GitHub’s degraded Git operations, then said their runtimes had returned to normal after mitigation. These are distinct service timelines across midnight, not a general Grok model outage. Original Bugbot/Cloud Agents notice.

Postmortem update for the older 30 October GitHub incident

GitHub updated its 30 October Actions, Pages and Webhooks notice on 3 November with a root-cause account. It attributes the disruption to a connection storm following cloud-network-gateway maintenance, amplified by aggressive Redis-client retries. It says removing affected resources restored the network path and that retry behavior is being improved. This is an older event’s explanation, not another November outage.

Descript: Avatars restored after a downstream model issue

Descript restored Avatar functionality at 01:56 on 3 November and closed the incident at 03:04. It first identified a downstream issue affecting models used by Avatars at 21:27 on 2 November. The affected component was Underlord – Video / Avatars. Descript’s timeline.

The provider did not name the downstream supplier or offer a workaround. The restoration update precedes the administrative resolution by about 68 minutes; neither should be confused with an independently established impact-start time. Other Descript features are not described as unavailable in this notice.

Cerebras and Fireworks: model-service recoveries

Cerebras says some models were temporarily unavailable from 19:52 on 3 November. By its 20:07 notice it had implemented a fix and was monitoring; it marked the incident resolved at 22:41. The notice does not identify the models or technical cause. Cerebras incident.

Fireworks serverless endpoints: separate resolved degradations

Fireworks resolved multiple model-specific performance incidents during the window. The longest published notice interval concerned GLM 5.3, from 3 November 11:47 to 4 November 07:52. Each row below links to a separate original notice. These endpoints are operated by Fireworks; they do not establish outages of the model developers’ own services.

The notices describe impaired performance and recovery, without a technical cause or a confirmed workaround. These are published notice times, not independently measured continuous downtime. Overlapping Kimi incidents must not be added together as a single duration.

Older issue still open: a separate 30 October Fireworks notice, first posted at 23:51 Istanbul, still lists deployment creation, dedicated training and batch inference as degraded. It has no newer update establishing recovery. Its age is preserved; it is not presented as a new 4 November incident.

Bolt: prompting remains degraded

Bolt’s Prompting component remained in the identified state at cutoff. The first notice was at 16:26 on 3 November; at 21:27 the provider said it had identified the issue and was working on a fix. No restoration time, cause or workaround was supplied. Bolt prompting incident.

This concerns Bolt’s prompting workflow. The notice does not establish an outage of its upstream model providers or every Bolt feature.

ChatGPT images: corroborated customer reports of reference inconsistency

Several distinct community authors report weaker adherence to reference images, character likeness and requested edits in ChatGPT. A 1 November post describes changes noticed from 29 October; independent 2–3 November posts describe similar failures in established workflows. These are customer-reported quality problems, not an acknowledged OpenAI outage. Initial account, independent reference-image account, 3 November account.

Counterevidence matters: one participant found improved photographic rendering but worse storyboard consistency; another reported differing results across accounts on 4 November. Mixed results, account variation.

A suggested workaround—shorter prompts and disabling memory—helped one author but failed for others. A participant also quoted a support reply giving prompting advice; that secondhand reply does not establish provider acknowledgement of a regression. Workaround discussion, failed attempts, reported support response. We cannot establish prevalence, a shared cause, an exact model version or a recovery date from this sample.

Sources

Original incident notices are linked beside every event and in the Fireworks table. Official OpenAI history, Claude incident history, Google AI Studio history and ElevenLabs history were also rechecked, alongside the wider creative-tool watchlist. Previously reported recoveries were deduplicated, including 3 November’s ChatGPT file-upload recovery. That recovery does not resolve separate image-quality, connector or account-specific complaints.

Public histories and accessible reviews do not cover every account, plan or region. Some review pages were inaccessible. No absence of a notice is treated as proof of uninterrupted service. For current releases and rollout changes, read today’s product news.

MUVO publishes these reports and offers competing creative tools. See our reporting method. Provider restoration claims and community observations are attributed rather than presented as MUVO testing.

Update timeline

• 4 November 2026, 11:54 Istanbul / 08:54 UTC: Research cutoff. Recorded the active ElevenLabs and Bolt notices, separate resolved media/API/coding incidents, an older Fireworks issue still open, GitHub’s postmortem update and corroborated ChatGPT image complaints with counterevidence.