AI incidents: Claude and OpenAI recoveries, Google London findings — 22 October 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

Anthropic reports recovery from connector errors and a separate Claude Fable 5 incident; OpenAI reports recovery from regional conversation errors and several narrower API/access incidents. Google has published a preliminary analysis of its earlier London disruption. These are separate events, not evidence of one shared outage.

The research window is 19 October at 12:01 through 22 October at 11:58 Europe/Istanbul (UTC+3). The table records provider acknowledgements and recovery notices, which do not necessarily mark the start and end of every customer’s impact.

Claude: connector errors and Fable 5 requests recover

Connectors in claude.ai and Cowork

Anthropic marked elevated connector errors resolved at 11:07 Istanbul on 22 October. The affected components were claude.ai and Claude Cowork. Investigation began at 06:50, and a deployed fix entered monitoring at 10:28. The notice does not name individual connectors, regions or a root cause. Original incident, official incident data.

The approximately 4 hours 17 minutes between investigation and resolution measures the public-notice interval. The source does not independently establish when customer impact began. It does not say failed connector actions were replayed.

Claude Fable 5

A separate notice reported elevated errors on Claude Fable 5 requests on 21 October, listing claude.ai, the Claude API, Claude Code and Cowork. Investigation was posted at 22:49 and resolution at 23:29: approximately 40 minutes between notices. Provider timeline.

The resolution text refers generically to other Claude models; the initiating notice specifically names Fable 5. We retain that narrower scope rather than infer an outage across every model. No cause, workaround or separate impact-start time is supplied.

OpenAI: regional conversation errors and narrower API incidents

ChatGPT Conversations and Responses API in India, Japan and Korea

OpenAI marked this regional incident resolved at 11:50 Istanbul on 22 October. The notice identifies India, Japan and Korea and lists ChatGPT Conversations and the Responses API. Investigation began at 05:18; mitigation and monitoring followed at 05:42. Original incident.

That is approximately 6 hours 32 minutes from acknowledgement to resolution, not a demonstrated continuous outage for every user in those countries. OpenAI gives no root cause or account-specific recovery procedure. Earlier cached versions of the notice still showed monitoring during research; the original page’s latest timestamped data records resolution before our cutoff.

FedRAMP Astra and Sol

OpenAI reports recovery from elevated errors affecting Astra and Sol in its FedRAMP service on 21 October. The sequence is unusual: an identified update at 16:42 precedes an investigating update at 16:51; monitoring follows at 19:46 and the resolution notice at 22:34. Official record.

The page’s component-status interval ends earlier, at 20:04, than its public resolution notice. We preserve both timestamps and do not turn that mismatch into a single outage-duration claim. This notice concerns FedRAMP, not all commercial OpenAI APIs. Cause and actual customer-impact boundaries remain unknown.

GPT-5.6 Luna requests

OpenAI resolved GPT-5.6 Luna errors affecting Chat Completions and Responses at 19:44 Istanbul on 21 October. Identification was at 18:34 and monitoring at 18:47, an approximately 1 hour 9 minute acknowledgement-to-resolution interval. No cause or workaround is given. Model-specific incident.

Authentication near Illinois on 20 October

OpenAI says some users near Illinois had authentication errors from 19:35 to 20:35 Istanbul on 20 October. Its notice covers API and ChatGPT login, plus Codex Web, API, CLI and VS Code. OpenAI attributes recovery to rolling back a configuration change. Provider notice.

Here, a one-hour impact interval is defensible because the provider explicitly supplies the start and end, originally 09:35–10:35 PDT. This is newly discovered coverage of an earlier event, not a new 22 October disruption.

Google Cloud: preliminary London findings revise the timeline

Google’s 22 October preliminary report changes the impact start to 09:00 Istanbul on 20 October and identifies a 6-hour-10-minute main recovery window ending at 15:10. Backup-generator control failures exhausted battery power in europe-west2-b; failover demand and cross-zone dependencies spread disruption. Speech-to-Text and Dialogflow CX are included in the expanded scope. Preliminary report.

The dashboard still ends the event at 17:00. That closure interval differs from the main impact window; neither proves simultaneous recovery for every workload. Google says over 99% of Compute Engine VMs were healthy by 15:10 and directs remaining cases to Support. Final analysis is pending.

We added dated corrections to the 20 October report and 21 October follow-up. Their original URLs and publication dates remain intact.

What affected users can check

Match the provider’s named component, region and model to the failed operation. Preserve the request identifier and timestamp for support. A provider’s recovery announcement does not prove a particular conversation, export or paid generation completed; check its state before starting another potentially billable operation.

Previously reported Work loading and Adobe-hosted generation problems have no newly verified general resolution in this scan. New isolated complaints about credits, cancellations and output quality remain unpublished leads because they lack matching independent evidence or relevant official corroboration.

Sources

• Anthropic: connector incident, Fable 5 incident, and public incident API.
• OpenAI: regional conversations/Responses, FedRAMP, Luna API, and Illinois authentication.
• Google: London report and timestamped incident feed.

Official notices establish only the stated scope. We did not reproduce failures inside private accounts. Inaccessible history, incomplete review feeds and undated claims remain coverage gaps. See our reporting method and separate product-news edition.

Update timeline

• 22 October 2026, 11:58 Istanbul / 08:58 UTC: Research cutoff. Added six provider-confirmed service events and Google’s preliminary follow-up; corrected the earlier London timeline with dated notes.