AI services, audio & video · Daily incidents

AI incidents: OpenAI and ElevenLabs recoveries, Firefly color reports — 13 October 2026

OpenAI has resolved the 13 October Codex and Work degradation. ElevenLabs resolves voice failures, while Firefly users report color shifts. Timelines and evidence explained.

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

Updated 14 October: the OpenAI investigation below is resolved; Firefly complaints remain customer-reported. The original edition recorded one investigation and three recoveries. The research window is 10 October at 11:53 to 13 October at 11:51 Europe/Istanbul (UTC+3). Events and updates may predate the edition date; dated recovery follow-ups appear below.

Affected workflow Latest provider status Useful event or notice time, Istanbul
OpenAI GPT-6.1 Sol in Codex and Work Resolved (13 Oct follow-up) Resolution notice 13 Oct 19:36
ElevenLabs Instant Voice Clone with noise removal Resolved Confirmed impact 13 Oct 06:25–07:44
ElevenLabs text-to-speech API Resolved Resolution notice 12 Oct 19:30; full impact interval not stated in the notice
OpenAI Compliance API Costs data Resolved; continuation of earlier coverage Resolution notice 12 Oct 19:02
Adobe Firefly Video color and brightness Customer-reported; no confirmed cause or fix Independent original accounts dated 10 and 12 Oct

Notice times are not automatically outage boundaries. See today’s separate AI news edition for launches and product updates.

OpenAI: GPT-6.1 Sol degradation in Codex and Work resolved

Update, 14 October: OpenAI marked this incident resolved on 13 October at 19:36 Istanbul, following monitoring at 12:30. Its initial acknowledgement was 11:10, with the model named at 11:18 and investigation continuing at 11:22. Original incident and recovery.

The affected components were ChatGPT Work, Codex Web, Codex API, CLI and the VS Code extension. Exact impact boundaries and root cause remain unspecified. A separate 14 October incident has a different ID; a common cause or continuous outage is not established.

ElevenLabs: voice-clone and text-to-speech incidents resolved

Instant Voice Clone: noise-removal requests failed for 79 minutes

ElevenLabs gives an explicit impact interval of 06:25–07:44 Istanbul on 13 October, equivalent to 03:25–04:44 UTC. That is 79 minutes. Some Voice Lab requests and the Voices API’s POST /v1/voices/add failed when background noise removal was enabled; requests without it were excluded. Official voice-clone notice.

The provider marks the event resolved. The page displays 03:25 UTC beside its retrospective notice, which also describes the later recovery; we therefore do not treat that display time as proof of when the recovery announcement was published. No root cause is given. This does not establish an outage of all voice generation.

Text-to-speech: provider attributes failures and latency to a traffic spike

A separate ElevenAPI Text to Speech notice is marked resolved at 19:30 Istanbul on 12 October. ElevenLabs attributes the transient error to heavy traffic and says it plans resilience improvements. Original TTS notice.

The written update gives no model or region list, full impact interval or customer workaround. We do not infer a duration or extend the incident to ElevenMusic, Dubbing or Image & Video.

OpenAI: delayed Compliance API Costs data has recovered

OpenAI marked its earlier Costs-data delay resolved at 19:02 Istanbul on 12 October. This updates the monitoring status in our 10 October and 11 October reports; it is not a new outage. Recovery notice.

The provider had excluded ChatGPT usage and API Platform requests from the reported impact. Its recovery claim concerns reporting freshness, not evidence of incorrect charges. The notice does not supply a precise original impact start.

Adobe Firefly Video: independent users report color and brightness shifts

Separate accounts dated 10 and 12 October describe image-to-video outputs changing exposure or color despite prompts asking for consistency. These qualify as corroborated customer reports, with no provider-confirmed model regression or platform-wide outage.

The 10 October brightness account says both the preview and exported video change brightness, and that prompt changes did not help. Later replies from distinct accounts describe similar symptoms. Repeated replies by the same person count once.

A separate 12 October report identifies Firefly Video, 1080p, 16:9, 24 fps, five seconds and start/end reference frames. The author reports darker orange tones across Chrome and Edge and says other models worked. This is their experience, not a controlled comparison by MUVO.

Adobe’s community manager requested the full prompt, output and browser. A community expert suggested lighting and reference-image prompts; the author disputed that prompting explained the change. Those replies do not confirm a fix. A further magenta-color complaint has less certain model attribution and is not needed to establish the two-account evidence threshold.

If reporting the same symptom, provide the selected model, browser, settings, prompt and before/after examples through Adobe’s support channel. No credit refund, universal workaround, common cause or frequency estimate is established by these threads. The accounts are public firsthand reports, not verified purchases or MUVO test results.

Earlier reports: Claude recovery and a Windows loading account

Update, 14 October: Anthropic resolved this reporting incident on 13 October at 21:00 Istanbul. See the recovery roundup.

In the previously covered ChatGPT loading thread, a 12 October, 15:20 Istanbul account adds that responses appeared on a phone while failing to appear in the Windows app. That is an anecdotal continuation, not a proven workaround or evidence that today’s Codex incident caused the earlier complaints. The earlier report now includes this update.

Sources

The scan covered major AI services and music, video and remix/stem tools. Isolated billing, capacity and quality complaints remain unpublished leads. Google’s Gemini API history and some other source histories were inaccessible or incomplete. A coverage gap proves neither an outage nor normal operation. Provider resolution does not verify every customer’s recovery. See our reporting method.

Update timeline