ElevenLabs Media Outage
The internet is broadly healthy today. AWS, Google Cloud, Azure, Azure DevOps, Cloudflare, GitHub, OpenAI, Meta, Stripe, Vercel, Heroku, Discord and Atlassian all report normal operations. There is no sign of an internet-wide event — no significant BGP leak, no submarine cable break, no DNS failure.
ACTIVE: ELEVENLABS MEDIA DELIVERY
ElevenLabs is investigating media that will not load across Studio, Audiobooks, Image & Video, Music and Dubbing. Both its ElevenAPI and ElevenCreative components are marked degraded. The incident opened at 08:49 UTC and, as of this report, has neither a cause nor a resolution. Source: https://status.elevenlabs.io/
ACTIVE, NARROW SCOPE
• Twilio has four open issues — three carrier-specific SMS delivery delays (ICE Costa Rica, Dukagjini Kosovo, Telefónica El Salvador) and one where recently upgraded accounts are still hitting trial-account restrictions. Twilio's Carrier Network and Developer Tools sections show degraded performance; the core APIs do not. Source: https://status.twilio.com/
• Slack has carried a minor incident since August 13: workspaces with custom data retention policies have trouble reaching historical messages. Messaging and workspace administration are flagged; login, files, huddles, search and apps are not. Source: https://slack-status.com/
• Akamai's edge delivery problems in India, open since August 22, are in monitoring after a fix. Akamai has ruled out the third-party provider it initially suspected. Source: https://www.akamaistatus.com/
• DigitalOcean is running extended Spaces maintenance in London, Singapore and Sydney, with brief interruptions possible. Planned work, not an outage.
THE DAY'S CROSS-PROVIDER STORY: AI MODEL ERRORS
Overnight, Anthropic logged elevated errors on requests to Claude models including Opus 5 and Fable 5, from 9:50pm to 12:36am PT. The same window showed up downstream at two other vendors: GitHub reported reduced Copilot availability for the Fable model "due to an upstream provider," and Cursor reported elevated Claude Opus 5 errors from 05:38 to 08:51 UTC. All three are resolved. It is a useful reminder that AI coding tools now share a small number of model backends, so one provider's bad hour becomes several products' bad hour.
RESOLVED IN THE LAST 72 HOURS
• Cloudflare: Durable Objects and D1 degraded in Hong Kong and Singapore, and elevated HTTP 5xx errors in Chicago. Both minor, both resolved earlier today. Maintenance continues in Ashburn, Montréal and Astana.
• Zoom: roughly 40 minutes of degraded meeting chat today, resolved.
• GitHub: on August 21, SSH Git operations fell below 95% success for about four minutes, and Copilot agent task creation failed intermittently. Git over HTTPS was unaffected.
• DigitalOcean: a significant interruption in the MKC1 region on August 19 affecting GPU Droplets is largely restored, with GPU node recovery continuing.
ONE SOURCING CAVEAT
AWS's public RSS feed is currently serving stale March 2026 entries, so today's AWS read comes from the Health Dashboard and secondary reporting rather than the feed. Neither showed an open broad event. Third-party monitors noted a short AWS warning window on August 23 that we could not corroborate against AWS's own dashboard, so we are not treating it as a confirmed incident.
Overall: incidents detected, none widespread. The internet is up.