The outage of Microsoft 365 was not an attack, and that’s what should concern businesses

In its reports on the global outage of Microsoft 365, the company outlines a cascading failure believed to be linked to a configuration change. This occurred while Microsoft was aggressively expanding its infrastructure and implementing voluntary departures. The incident raises concerns about the reliability of a backbone crucial for global communication and transactions—elements that are not fully addressed by Microsoft’s service credits or European commitments.

The cascading failures among major global providers have cast doubt on the reliability promised by cloud services. Microsoft’s Exchange Online, Teams, SharePoint, and other dependent services rely on shared databases for connectivity and identity; thus, when email systems fail, the fallout affects multiple channels. This was evident starting the evening of November 12, Paris time, when Microsoft first reported access issues with Exchange Online in the Americas, later acknowledging that Microsoft 365 connectivity could be affected “in all regions.”

By November 13, Microsoft noted improvements in North America and a sustained recovery in much of Europe, with some encouraging signs in the Asia-Pacific region, although no timeline for returning to normal service was provided. The company extended the retention period for pending emails to 48 hours to prevent loss of delayed communications. Microsoft only declared service fully restored for its last customers on November 14, nearly two days after the outage began. On that Monday, Microsoft issued a preliminary post-incident report, limited to its administrators.

Cascading Failure: Neither Attack Nor Overload

Microsoft’s updates on its service status help eliminate various hypotheses due to the absence of public technical explanations. A saturation attack would imply a surge in requests, but Microsoft attributed the high load to servers becoming unavailable, which then redirected their load to others without a corresponding increase in traffic. A compromised system would show signs of containment, such as the rotation of identifiers or isolation of accounts. Instead, reports indicated actions like restarts, traffic rerouting, and limitations. Microsoft stated that it had “no indication” of malicious activity. The dismissal of these two possibilities is based solely on the company’s statements and corroborated by nothing it published.

However, some servers were failing automated health checks, which determine whether they should remain operational, and Microsoft suspected the component responsible for managing those checks. A central process appeared overwhelmed, leading those servers that were restarted to fail again as soon as they received traffic. Experts in distributed systems refer to this situation as a metastable failure. The initial cause may have disappeared, but each restored capacity attracted enough load to cause it to fail again. This mechanism potentially explains the duration of the outage better than any extraordinary cause, resembling the October 2025 incident with Azure Front Door, where removed nodes processed their traffic onto others until they became overloaded.

A Configuration Change Amid Ongoing Infrastructure Work

On November 12, Microsoft stated it had identified “no recent changes” correlated with the beginning of the incident. Nevertheless, the company reconstructed some servers “with a previous configuration” to check if they would stop disconnecting and paused all deployments. This sequence suggests a fourth hypothesis: an incident that occurred in a conducive environment. A flaw introduced during a previous modification, only revealed when interacting with another change, is inherently difficult to trace to its original cause. The preliminary report is expected to identify the specific change involved, but it remains limited to client administrators.

Microsoft spent $39.4 billion on capital expenditures in the quarter ending September, a significant increase from $19.4 billion a year prior, to expand its cloud and AI infrastructure. This rapid pace of growth forms a crucial part of the current context. The more capacity increases, the more latent defects can circulate, and the operational teams are often working in an ever-changing environment. The second factor is staffing. Following a voluntary retirement program earlier in the year, Microsoft opened a new voluntary departure offer on October 26 to U.S., Canadian, U.K., and Indian teams, including senior executives. These departures take away operational knowledge that is seldom documented, based on workarounds known to engineers but absent from official records. While a direct link to the outage is not established, this may serve as a complicating factor that only the final report can confirm or rule out.

Europe Affected, Beyond European Commitments

Europe was not spared from the outage that originated in the Americas. The company has made numerous commitments to its European clients, including maintaining the storage and processing of Microsoft 365 data within the European Union and the European Free Trade Association since February 2025. In April 2025, it also promised to work with European partners to ensure operational continuity should it be compelled to cease operations in Europe.

These commitments relate to data locations and legal jurisdiction, and none address the technical interdependencies between regions, which ultimately failed. A shared layer collapsed, causing Europe to be affected alongside the rest of the world. The promised isolation is legal, not operational. Sovereignty addresses the risk of a political decision but does not protect against a failed configuration change—something that can happen to any provider, European or otherwise.

Related Risk, Cost Borne by Clients

When a platform of this magnitude fails, thousands of organizations lose their channels for communication and validation, leaving them with no fallback solutions. The scale of the provider and the extent of the outage create systemic implications affecting all sectors and user classes. Regulators have acknowledged this in the financial sector. Since November 2025, Microsoft Ireland Operations Limited has been listed as a critical service provider under the Digital Operational Resilience Act (DORA), placing Microsoft’s Irish subsidiary under the direct oversight of European regulatory authorities.

European authorities thus monitor the risk management of the subsidiary without compensating its clients. Their standard contractual recourse remains service credits, calculated per client per month based solely on the fees for the failed service, which administrators must request. In the example published by the company, monthly availability below 95% entitles clients to a credit equal to that month’s subscription. A client without service for two full days, equating to nearly 7% of a thirty-day month, might recover the cost of one month’s subscription, while the operational losses they incurred remain their responsibility. Ultimately, the cost of the outage lies with the client, and there is little financial incentive for the provider to slow down changes.

The company must first assess the cost of a day without messaging or collaboration, compare it to the recoverable credit and insurance guarantees against third-party failures, and then request post-incident reports. The lesson speaks to accounting before politics. Primarily, it emphasizes the need for organizations to maintain independent crisis channels and functions that must not falter alongside those of Microsoft.

Scroll to Top