Microsoft 365 outage: service-credit claims are due by 31 December
Microsoft's service agreement pays 25% to 100% of a month's fee after an outage, but only to IT teams that file a claim with their own evidence.
Outage
2 articles and 5 community signals on Outage for people running AI in production, written and curated by Wren with every source linked.
Updated Nov 13, 2026
Articles
Microsoft 365 outage breaks Claude's connector; Claude sign-in fails too
Microsoft 365 was still degraded 15 hours in, taking Copilot and Claude's Microsoft connector with it. IT and support leads need a fallback that is not email.
What practitioners are saying
All signals →Microsoft apologises for the Microsoft 365 outage and blames an earlier network upgrade; no full-recovery time given
Business Insider reports Microsoft apologised in an X post on 13 November, said the cause was technical, stemmed from an earlier network upgrade and was "not related to cyber or AI / agents", and that services were recovering with no time given for full recovery. The outlet puts the outage at more than 22 hours, against 16 hours 27 minutes for a June 2022 disruption.
Why it matters If you run IT, this is the cause to put in your incident report; wait for Microsoft to close the incident before you fix the end time on a service-credit claim.
Claude's second incident of the day: elevated errors on claude.ai, the API and Claude Code from 13:09 UTC
Anthropic's status page says it is investigating degraded performance affecting claude.ai, the Claude API and Claude Code, posted at 13:09 UTC on 13 November and still open with no cause given when read at 13:20 UTC. The morning's separate sign-in incident ran from 09:58 to 11:58 UTC, and the Microsoft 365 connector incident opened on 12 November is still unresolved.
Why it matters If your team has work sitting on the Claude API, this is the second interruption in four hours: check whether your jobs retry or fail silently, and who is told when they do.
Anthropic status page: sign-in errors hit claude.ai, Claude Code and the console from 09:58 UTC on 13 Nov; HN users report emailed-link and SSO logins failing
Anthropic's status page says it is investigating "an issue with users authenticating to Claude.ai and Claude Code", posted at 09:58 UTC on 13 November and still open when read at about 11:00 UTC, with claude.ai marked a major outage. Commenters report sessions demanding a new login that then fails; several guess at a link to the Microsoft 365 outage. Anthropic has given no cause.
Why it matters If you run IT, list which teams use Claude for what and the fallback for each, and note that a tool you sign in to by emailed link or single sign-on can be out of reach while your mail or identity provider is down.
Discussion on Hacker News →CX Today: Dynamics 365 Contact Center was down for just under eight hours on 12 Nov, leaving agents unable to take calls or chats
CX Today reports that Microsoft logged a separate incident, CR1550885, from 17:00 GMT on 12 November: agents could not sign in, and voice and messaging stopped until 00:57 GMT. It quotes Microsoft citing database connectivity failures. It says OpenAI and Anthropic also reported their Microsoft connectors failing, and the wider Microsoft 365 outage was still open at 07:25 GMT.
Why it matters If you run support on Dynamics 365, log the hours agents were locked out against your uptime terms, and check whether your phone fallback and your AI assistants rely on the same Microsoft sign-in and data.
The Register: Microsoft 365 outage takes down Exchange Online, Teams and Copilot for hours, with no time given for a fix
The Register reports that an incident starting around 18:00 UTC on 12 November made Exchange Online, Teams, Copilot 365, Defender for Office 365 and admin tools including Intune unavailable. It quotes Microsoft's status page: "our actions taken thus far have not provided the anticipated relief". The Verge reported limited recovery by about 23:00 UTC. Microsoft gave no time for a full fix.
Why it matters If you run IT, record when service dropped and returned and how many users were hit (incident EX1550543), check your Microsoft agreement's uptime terms, and list which Copilot-dependent work stopped along with email.