August 2026 AI Tools Outage & Uptime Report

August 2026 saw 26 major incidents totalling 4,153 minutes across 6 providers. GitHub and Anthropic logged 8 each and Cursor 7 — together the bulk of the month.

GitHub was the least stable provider. A GitHub Actions incident ran about 10h42m from August 7 00:22 (JST); “intermittent failures creating agent tasks” (Copilot’s coding agent) lasted about 9h54m from August 20 23:43; and a GitHub.com incident ran about 7h36m from August 17 22:40 — three long incidents at the highest severity, leaving GitHub with the lowest uptime of any provider at 95.16%. August 27 added incidents in Actions (about 2h50m) and in Copilot’s AI model providers (about 2h08m).

Anthropic’s eight incidents included elevated errors across multiple models for about 3h24m from August 24 14:06, and elevated errors in Claude Code and Cowork for about three hours from August 29 02:22. A disruption across Claude services at the highest severity (36 minutes) was also recorded at 06:58 on August 17.

Cursor’s notable incidents were a degradation of Automations, Cloud Agents and the IDE for about four hours from August 13 23:46, and a GitHub-caused degradation on August 17 (about six hours).

OpenAI, by contrast, had zero major incidents in August (16 minor degradations). Other incidents included a partial outage of Replicate’s A100 GPUs (about three hours from August 27 06:17) and a Perplexity API issue (about an hour from August 13 09:45).

This month clearly recorded how long it takes an upstream outage to reach downstream services. GitHub.com’s incident began at 22:40 on August 17, and 54 minutes later Cursor announced that a GitHub degradation was affecting some Cursor services. On August 13, too, Cursor announced degradation of Cloud Agents and more 13 minutes after Anthropic announced errors for Mythos 5, Fable 5 and Sonnet 5.

Cascades also start at the model providers. During Anthropic’s August 24 incident, GitHub announced “elevated errors on Fable 5 due to upstream provider”. Cursor’s “Fable 5 unavailable” on the morning of August 20 (about 2h41m) also traced back to the model provider. AI tools depend on several companies’ models and infrastructure, so one company’s outage often surfaces under another company’s name.

Minor degradations (latency-level) numbered 81 — Cursor 20, GitHub 18, OpenAI 16 and Anthropic 14 — about three times the 26 major incidents.

The month total of 4,153 minutes is above June (2,759) and less than half of July (8,901). July included the resolution of a deliberate model suspension, so a direct comparison is imperfect, but August was a relatively calm month without any single large-scale outage.

Incident counts and downtime include only incidents on each provider’s official status page that were both “major or worse (effectively unusable)” and already resolved (calendar month, Japan time). Minor degradations such as latency, and unresolved ongoing items, are excluded from the totals so the figures aren’t inflated beyond real impact.

There were no unresolved major incidents at the end of August.

Months are counted in Japan time. A Cursor incident that began on July 31 UTC (04:50 on August 1 JST) is therefore counted in August.

Uptime is our own estimate based on that major-outage time — not official SLA figures. Providers that split their services into many separately reported components, such as GitHub, tend to look worse in the numbers.

Each service’s “outage history” page lists individual incidents with title, severity, duration and the official link.