The warning and the Microsoft 365 case
Satya Nadella, Microsoft’s CEO, warned of a possible increase in digital service blackouts caused by artificial intelligence. Speaking on November 18 at the DealBook Summit in New York, he explained that AI tools identify vulnerabilities so quickly that they force companies to deploy fixes almost immediately, when the flaws are already being exploited by attackers. In the past, he added, companies had weeks to test an update; now that window can shrink drastically.
The statement comes a few days after the global Microsoft 365 outage, which between November 12 and 13 especially compromised access to Exchange Online and some related functions. Microsoft attributed the incident to a software defect and a chain of infrastructure problems, ruling out both AI and a cyberattack. The distinction matters: Nadella’s warning concerns the future, not the diagnosis of that blackout.
The mechanism of the outage
According to Microsoft’s preliminary report of November 15, an update rolled out starting in July on Microsoft 365 network gateways had introduced a bug that remained latent for more than three months. Under certain load conditions, an internal Windows process contended for shared resources, saturating CPU and memory. On November 12, about fifteen hours before the disruption, some Teams traffic was shifted to the gateway that also served Exchange Online; during the same period, nearly 14% of the capacity of those gateways was unavailable because of a separate problem.
The combination surfaced the defect while automated systems moved requests from struggling servers to still-healthy ones, overloading the latter as well. At the worst point, more than 80% of the servers on the affected gateway were not responding, a figure that does not concern Microsoft’s entire infrastructure. The outage lasted almost 24 hours, from 16:20 UTC on November 12 to 16:10 UTC on November 13. Testing had passed the faulty update because laboratory tests did not reproduce the critical condition.
Data on vulnerabilities and limits of the forecast
Nadella’s warning finds partial corroboration in the analysis published on September 30, 2026, by the Google Threat Intelligence Group. Disclosed vulnerabilities rose from 5,045 in January to 10,740 in August 2026; those observed under active exploitation rose on average from 10.5 per month in 2025 to 18 in the January–August 2026 period. Google warns, however, that the growth also depends on changes in how CVE identifiers are assigned and that only a small fraction of disclosed vulnerabilities was observed under active exploitation, about 0.23% in the period analyzed in 2026.
Nadella did not provide an estimate of how many additional blackouts AI might cause. His remains a forecast about the possible imbalance between the speed of attacks, patch security, and system robustness, not a demonstration of an already established statistical relationship. He also evoked the fear of the Millennium Bug, while noting, however, that it was a bug tied to a specific date, whereas the race among AI models, discovery of flaws, attacks, and defenses has no deadline.
