On Friday, November 13, 2026, Microsoft apologized for an outage of Microsoft 365, its online office suite. For more than a day, some users could not reach Outlook email, the calendar in the Teams meeting app or the Copilot AI assistant. In a post on X, Microsoft called it “a technical issue.” It said the incident was not related to a cyberattack, AI or AI agents, which are programs that carry out a chain of tasks on their own.
Microsoft blamed a problem during an earlier upgrade of its network capacity and promised a detailed report. Service came back gradually on Friday. On Saturday, November 14, the company’s status page said service had been restored for the last affected customers and regions, the British newspaper The Guardian reported.
A 30-hour outage by Microsoft’s own count
Microsoft’s first post about the outage on X came at 6:54 p.m. Paris time on Thursday, November 12, from its Microsoft 365 Status account. That account, which reports on service health, said Microsoft was investigating problems reaching Exchange Online mailboxes “in the Americas.” Exchange Online is Microsoft’s business email service. Thirteen minutes later, the account said users in any region could be affected.
That evening, Microsoft paused all update deployments and restarted the affected servers. At 8:31 p.m., it explained why its systems were under heavy load. A number of servers had become unavailable without warning, and their work was falling on the machines still running. Microsoft ruled out a rise in traffic.
At 4:13 a.m. on Friday, Microsoft admitted that its overnight changes were not working as expected. At 10:24 a.m., it reported early signs that service was stabilizing. It also extended the time its servers hold delayed email to 48 hours, so messages would not be lost.
Microsoft’s final summary is in notice MO1550821, the incident record in the Microsoft 365 admin center. The admin center is the part reserved for customers’ IT staff. The summary puts the outage at about 30 hours, from 11:05 a.m. on Thursday to 5:10 p.m. on Friday, Paris time.
That 11:05 a.m. start is almost eight hours earlier than Microsoft’s first post on X. On Thursday, the same notice still put the start at about 7:25 p.m., the tech site Windows Latest reported. The Guardian says visible problems began at about 7 p.m.
Copilot, Teams and admin tools were hit too
Microsoft’s notice lists about 20 affected services. It says only features that run through Exchange Online were disrupted. Teams calls, meetings and group chats mostly kept working, but the Teams calendar, invitations and meeting scheduling failed.
The outage also reached admin and security tools. The notice names the Microsoft 365 admin center, the Defender security portal, the Intune device-management tool and the Purview data-protection tool. Files in SharePoint, Microsoft’s shared document service, stayed available, but sharing them with people outside the organization could fail.
Copilot is on the list too. The notice says some users might not be able to open Copilot Chat, the assistant’s chat window, or might get no answer. That could happen on the web and in the desktop apps. The assistant could also lose access to work email, calendars and files.
Rival AI assistants also depend on the services whose data they read. That is true of Google’s Gemini once it is linked to Gmail, Docs and Drive, and of Anthropic’s Claude through its connectors for Gmail, Drive and Google Calendar.
Microsoft rules out AI, as Amazon did in February
In a November 14 story on Microsoft’s denial, the French tech site Numerama reported that theories were circulating on social media. Some blamed code written by AI. Others blamed a network setting approved without proper human review. As early as Thursday night, one Reddit user asked what the odds were that rogue AI agents would be blamed.
Numerama points out that Microsoft builds Copilot into all of its products. In April 2025, Microsoft chief executive Satya Nadella estimated that AI wrote 20% to 30% of the code in the company’s repositories.
On October 19, 2026, Microsoft itself published two blog posts that describe AI agents as insider risks. The posts say an agent should be watched like an employee with too many permissions.
Amazon faced the same suspicion in February 2026, over an outage of its own. That month, the Financial Times reported that Kiro, Amazon’s AI coding tool, had decided to “delete and recreate the environment.” The British newspaper said that decision led to a 13-hour interruption of a service at AWS, Amazon’s cloud business, in mid-December 2025.
In a blog post cited by the tech site GeekWire, Amazon replied that the cause was a misconfigured role, meaning badly set access rights. It called this “the same issue that could occur with any developer tool (AI powered or not) or manual action.” Amazon said the disruption affected a single cost-tracking tool in one region of mainland China.
The impact in France and across Europe
In France, 82% of companies with an office suite use Microsoft 365. The figure comes from the Observatoire de l’intranet Arctus 2025, a study cited by the French newspaper Le Monde. Le Monde reports that many employees could not use Outlook on Friday morning. Even so, it describes the impact on companies as limited, because by its account the outage lasted only from Thursday to midday Friday.
Government agencies in the Netherlands, Belgium and France reported problems, according to The Guardian. One French town asked residents to call instead of sending email. In the UK, hospitals in the NHS, the country’s public health service, went back to paper records. Staff at a Manchester court scrambled to find a fax machine.
Downdetector, a site that counts outage reports from users, received reports from more than 70 countries and territories.
Microsoft’s contract offers credits for downtime
Windows Latest reviewed the November 2026 version of Microsoft’s service level agreement. That contract sets the share of time Microsoft promises its services will be available to business customers. If Exchange Online’s monthly availability falls below 99.9%, the site says, the customer is entitled to a credit of 25% of the fees for that service alone. The credit rises to 50% below 99% availability and to 100% below 95%.
An earlier version of the agreement, cited by the site, bases the calculation on both how long the service was down and how many users were affected. Customers have to claim the credit themselves, no later than one month after the end of the billing month in question.
Questions Microsoft has yet to answer
The cause Microsoft gives is still preliminary. It traces the incident to a problem during an earlier upgrade of its network capacity. That problem surfaced after a recent change to the network. Microsoft has not said what that change was, or why its first fixes failed to hold overnight.
Its explanations also shifted during the outage. On Thursday night, it described unavailable servers and an overwhelmed core process. At around 2:12 a.m. on Friday, notice MO1550821 said Microsoft had begun rolling back a recent network configuration change “that correlates with the start of impact.” On Friday afternoon, the company named the network as the cause.
Steven Murdoch, a professor of security engineering at University College London, told The Guardian that an outage of this size usually results from “some surprising interaction between systems.” He said each system is normally quite well understood, but the interactions between them are complex.
The official timeline raises questions too. The final summary ends the incident at 5:10 p.m. on Friday. Yet at 9:05 p.m., almost four hours later, Microsoft still said it had no estimate for full recovery. It also has not explained why the summary starts the incident as early as 11:05 a.m.
In its final summary, Microsoft promises a preliminary post-incident report within two business days and a final one within five. Until then, Microsoft’s claim that AI played no part rests on its word alone. The company has published no technical evidence to back it up.
What it means
The outage shows how much Copilot depends on the rest of the suite. The assistant works with a user’s email, calendar and files, and it loses access to them when Exchange Online goes down. Microsoft sells Copilot as an everyday work tool, but it fails along with the services it queries.
The main risk is relying on a single provider. Email, meetings, files and AI all come from Microsoft and run partly on the same infrastructure. Given the Arctus figure of 82%, an outage like this one can hit most French companies with an office suite at the same moment.
The AI denial also matters commercially. Like Amazon in February, Microsoft chose to rule out AI publicly as the cause of an outage. That theory was circulating on social media at the time, according to Numerama. Both companies sell AI tools to businesses, and the suspicion hits their sales pitch directly.
The preliminary report, due in the coming days, will need to name the network change that triggered the outage. It will also need to say whether a person or an automated system applied that change, and why recovery took more than a day. That report will be the first chance to check Microsoft’s claim about AI.
Sources
- “We are deeply sorry for the disruption this incident has caused our customers and partners”, Microsoft on X, November 13, 2026
- Microsoft 365 Status posts on incidents EX1550543 and MO1550821, Microsoft on X, November 12 and 13, 2026
- MO1550821: Users may be unable to connect to multiple Microsoft 365 services, Microsoft 365 notice as captured by 365Fanatic, November 14, 2026
- Microsoft says services fully restored after global outage hits governments, The Guardian, November 14, 2026
- La panne de Microsoft alerte entreprises et salariés sur leur dépendance aux outils numériques, Le Monde, November 13, 2026
- Panne géante : Microsoft balaie la théorie d’un agent IA qui aurait dérapé, Numerama, November 14, 2026
- Amazon pushes back on Financial Times report blaming AI coding tools for AWS outages, GeekWire, February 20, 2026
- Microsoft 365 outage: you may get money off your bill if you use it for work, Windows Latest, November 13, 2026
Header image: MACCUS chart, data from Microsoft and Microsoft 365 Status on X, the MO1550821 notice and The Guardian, November 12-14, 2026
