AI incidents: Claude Gmail recovery, Sonnet errors and Google’s London postmortem — 28 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

Claude’s Gmail connector, two Sonnet error episodes and OpenAI’s Sol API degradation are resolved in the official notices. Google has published a full explanation of its earlier London outage. Two independent Runway reviews also describe requested edits failing to appear while credits are consumed; those remain customer reports.

The coverage window is 25 October at 12:24 through 28 October at 12:03 Europe/Istanbul (UTC+3). Older events are included only for material follow-up. Times below distinguish reported customer impact from provider notice times.

Claude: Gmail connector recovery and two Sonnet 5.5 incidents

Gmail search and reading recovered on 28 October

Anthropic reports Gmail connector errors from 02:00 to 08:29 Istanbul on 28 October: 6 hours 29 minutes of stated impact. Searching and reading email were affected in claude.ai, Claude Code and Claude Cowork; the notice says other connectors were unaffected. Official Gmail incident.

The first acknowledgement was at 06:59, and the resolution was posted at 08:48. The duration uses Anthropic’s explicit impact boundaries, not those announcement times. No root cause or workaround is supplied. Provider recovery does not prove that every previously failed request or individual account has recovered.

Sonnet 5.5 had two separate error episodes on 27 October

The first Sonnet 5.5 notice was acknowledged at 12:14 Istanbul, moved to monitoring at 12:38 and resolved at 12:56. Actual impact boundaries were not supplied, so this is a notice sequence rather than a measured outage duration. First official incident.

A second incident has an explicit 35-minute impact interval, 17:42–18:17 Istanbul. Anthropic acknowledged it at 18:09 and posted resolution at 18:50. Both notices concern elevated errors for the named model and list claude.ai, Claude API, Claude Code and Claude Cowork; neither establishes that every model or user failed. The causes are unspecified. Second official incident.

OpenAI: Sol API recovery and a correction to hosted-review status

OpenAI resolved its “6.1 Sol Degraded Model Performance” incident on 27 October at 19:20 Istanbul. The investigation began at 18:02, with mitigation and monitoring at 18:40. The named components are API Chat Completions, Responses and Batch. Official Sol incident.

The notice does not describe the precise performance symptom, root cause or actual impact-start time. It therefore cannot support a quantified outage duration or a claim that all ChatGPT services were affected.

Correction to yesterday’s edition: the hosted-review notice now records resolution at 11:30 Istanbul on 27 October. Our earlier retrieval supported “monitoring”; the recorded resolution time predates that edition’s cutoff. This is an updated account of the same event, not a new recovery on 28 October. The scope remains most hosted Code Reviews and Security Reviews on private GitHub repositories. A dated correction appears in the 27 October report. Original hosted-review notice.

Google Cloud: London postmortem identifies power-control failure

Google attributes the 20 October London disruption to an external cable fault and generator-control misconfiguration. With one power feed isolated for maintenance, the controller incorrectly sensed usable power; generators did not start before battery reserves ran down. The affected zone was europe-west2-b, with regional failover pressure. The named services include Speech-to-Text and Dialogflow CX, not a Gemini-wide outage. Full provider report.

Google now gives 09:00–15:08 Istanbul as the main impact interval, 6 hours 8 minutes, revising the preliminary endpoint. Controller changes were completed on 26 October; further electrical repairs and validation remain planned. We added the final timeline to the 20 October report. This is a postmortem, not a fresh outage.

Runway: two customers report edits not taking effect while credits are used

Two independent reviews with experience dates of 26 and 27 October describe requested video changes not appearing after repeated attempts. The 27 October reviewer explicitly says credits were consumed; the 26 October account describes repeated unchanged results and an unsuccessful refund request. 27 October account, 26 October account, original review listing.

These are customer-reported output and credit concerns, not a provider-confirmed outage or proven billing defect. The accounts do not identify enough model, plan or input details to establish a common technical cause. No provider reply was visible for these entries. MUVO has not reproduced the results; two reviews do not establish prevalence. An unrelated single-review Topaz credit allegation remains unpublished pending corroboration.

Follow-up checks and practical limits

The existing ChatGPT scheduled-action thread received another same-author report: SharePoint copying works, while scheduled replacement still encounters a file-reference block. It adds no general resolution or new independent outage evidence. Check the destination result of a scheduled run, rather than relying only on its completion message. Dated follow-up.

ElevenLabs’ accessible history and older incident write-up pages supplied no newly verified recovery or completed postmortem. Other previously resolved official notices did not establish a fresh recurrence. Restricted review sites, sparse specialist histories and tenant-only status views remain coverage gaps; a green dashboard alone was not treated as proof that no earlier disruption occurred.

Sources

• Anthropic: Gmail connector, first Sonnet event, second Sonnet event.
• OpenAI: Sol API, hosted reviews, scheduled-action follow-up.
• Google: London postmortem.
• Runway: 27 October review, 26 October review.

MUVO publishes these reports and offers competing creative tools. Read our reporting method and separate product-news edition.

Update timeline

• 28 October 2026, 12:03 Istanbul / 09:03 UTC: Research cutoff. Added Claude and Sol incident notices, the hosted-review correction, London postmortem and corroborated Runway customer reports. Added dated follow-ups to the 20 and 27 October articles, preserving their URLs and original publication dates.