GitHub Copilot and Claude Recover After Overlapping AI Model Outages

GitHub Copilot and Anthropic’s Claude services have recovered after overlapping model disruptions early Monday. The official records confirm the service trouble and the recovery—but they do not prove that both incidents had the same technical cause.

If your AI coding assistant acted like it needed coffee before the rest of us, this was not necessarily your computer’s fault. GitHub reported degraded availability for the Fable 5 model in Copilot products and IDE surfaces. Anthropic separately reported elevated errors affecting Fable 5 and other Claude models, including direct Claude services.

What happened

GitHub opened a major incident at 3:12 a.m. EDT after reports of degraded availability involving its Copilot AI Model Providers component. Its update said Fable 5 was affected in Copilot products and IDE surfaces because of an issue with an upstream model provider. While engineers worked on the issue, GitHub advised users to select another model or use Auto.

Anthropic’s separate major incident involved elevated errors across multiple Claude models and partial outages on Claude.ai, the Claude API, Claude Code, and Claude Cowork. Its updates named Fable 5 among the affected models. Anthropic later said errors had stabilized on Opus 5 and Fable 5 while work continued on success rates for all affected models.

Recovery timeline

1. 12:50 a.m. EDT — Anthropic later identified this as the beginning of the period when users saw elevated errors on requests to Claude models.
2. 1:06 a.m. EDT — Anthropic’s public incident record opened, initially naming Mythos 5, Fable 5, Opus 5, and Opus 4.8.
3. 3:12 a.m. EDT — GitHub opened its Copilot provider incident and reported degraded Fable 5 availability.
4. 3:36 a.m. EDT — Anthropic’s final update says the elevated-error period ended.
5. 3:58 a.m. EDT — GitHub marked its incident resolved and said a detailed root-cause analysis would be shared when available.
6. 4:30 a.m. EDT — Anthropic marked its incident resolved.

What users may have noticed

For GitHub Copilot users who selected Fable 5, the practical symptom was degraded model availability or failed requests inside Copilot products and IDE integrations. Anthropic customers may have seen errors in Claude.ai, API-backed workflows, Claude Code, or Cowork. Neither vendor said that every model, every user, every plan, or every region was affected.

That distinction matters when troubleshooting. A vendor-side model incident can look like a broken IDE extension, an account problem, a bad local network, or a billing issue. Reinstalling software during a live provider incident can add confusion without fixing the upstream service.

What The IT Guys recommends after recovery

1. Confirm the official status first. Check the vendor incident page before changing a workstation, IDE, extension, account, or firewall.
2. Save work locally. Keep prompts, code changes, drafts, and unsent material outside the active AI session when the work matters.
3. Use another approved model when practical. GitHub’s incident-specific workaround was to select another model or Auto. Follow your organization’s privacy and data-handling rules when switching providers.
4. Make retries safe. A timeout does not always prove that nothing happened. Automations should use bounded retries, idempotency keys, reconciliation, and human review for consequential actions.
5. Reconcile downstream work. Verify generated files, commits, tickets, messages, and records before assuming a failed-looking request produced no result.
6. Keep a manual fallback. Important development and business workflows should still have documentation, local tools, and a path that does not depend on one model endpoint.

What remains unknown

GitHub has not yet published the promised detailed root-cause analysis. Anthropic said it identified the cause while the incident was active, but its public updates did not disclose that technical cause. The status records also do not provide affected-user percentages, geographic scope, plan-by-plan impact, data-loss evidence, or a verified connection between the two incidents.

In other words: the services recovered, the overlap was real, and the exact causal relationship remains unproven. That is less dramatic than “everything went down everywhere,” but it is much more useful for deciding what to fix—and what not to touch.