The same roundup covers two enterprise AI announcements that matter to Microsoft shops. Cisco is connecting AI security findings to ServiceNow remediation, and IBM and Microsoft have launched an agent platform called Imagine Enterprise. Below, each incident is covered separately, followed by what Windows and Microsoft 365 administrators can take from the week.
Microsoft 365: one Exchange Online fault, many affected services
For WindowsForum readers this was the biggest incident. It started on Thursday, November 12, 2026. TechRadar's live coverage quoted Microsoft 365 Status acknowledging that users in the Americas were having trouble reaching Exchange Online mailboxes "through multiple connection methods." The issue was tracked under EX1550543 in the Microsoft 365 admin center, with a broader incident under MO1550821.
Microsoft's updates, as relayed by TechRadar, showed a messy recovery:
- Servers were failing automated health checks and dropping out of service. Microsoft tested a configuration change and restarted affected servers in parallel.
- As a precaution, Microsoft paused all deployments and configuration changes while it looked for a link between recent changes and the outage.
- It said the heavy load did not come from more traffic. A large number of servers had become unavailable unexpectedly, which pushed their load onto the infrastructure that was still running.
- The failed health checks appeared to come from the component that processes server health. Microsoft validated an alternative health-check method, rolled it out more widely and set up automated recovery to detect and restart unresponsive servers.
At its peak, TechRadar reported more than 131,000 Downdetector reports for Outlook. Recovery continued into Friday, with Microsoft describing improvements in mailbox access and delayed email processing and warning that intermittent problems might persist. Microsoft's final public explanation said the issue was not related to cyber activity, AI or agents. It traced the outage to a problem introduced during an earlier network-capacity upgrade that surfaced after a later network change. Microsoft said it restored service gradually and checked stability at each stage, and it has promised a post-incident report.
That report hasn't been published yet, so treat the cause as Microsoft's high-level account during the incident, not a full root-cause analysis. It's also notable that Microsoft initially said it had found no recent change correlated with the start of impact, then later blamed an earlier upgrade combined with a recent network change. The post-incident report should show how those two statements fit together.
Why this matters for admins: Exchange Online is a dependency for much more than Outlook. Calendars and any feature that relies on mailbox data failed along with email. TechRadar also reported that Microsoft's public status page wouldn't load during part of the outage, and Microsoft directed customers to an alternative status channel if the admin center was unavailable. If your incident-communication plan assumes the admin center and Teams will be up, it may not hold during a large Microsoft 365 outage.
Workday: when disaster recovery causes the problem
Workday's incident happened the next day. Florida Atlantic University's IT status page posted Workday's notice on the evening of November 13. Workday had identified a problem in its disaster-recovery environment that produced inconsistent data in production, and as a precaution it put all WD1 Production tenants into maintenance. The university warned that Workday Production was unavailable and that scheduled jobs, integrations and other automated processes might fail.
According to UC Today, the DR services unexpectedly connected to production databases, and the outage disrupted timekeeping, payroll and benefits. The university's notice confirms the DR-to-production data problem but doesn't describe the connection mechanism, so that detail rests on UC Today's reporting.
The university's next morning update said Workday expected reindexing and tenant release for most customers to finish by 11 a.m. Eastern. Timing varied by tenant, and customers facing longer delays would be contacted. By 10:30 a.m. on November 14, the university reported that Workday Production was available again. That is one institution's experience. The sources don't support saying that every tenant recovered at once, or that any data was permanently lost.
The lesson is uncomfortable. Most recovery plans assume the DR environment sits idle until it's needed. Here it appears to have damaged production. If your DR runbooks don't consider how a failover environment could write to live systems, they're incomplete.
JPMorgan Chase: an on-premises bug that affected other banks
UC Today reports that JPMorgan Chase blamed a software bug in an on-premises hardware system for a temporary technical issue and said it was not cyber-related. Chase customers had trouble with online accounts. People at other banks trying to transact with Chase also saw problems, and Zelle payments involving affected institutions were delayed.
That distinction matters. The reporting describes delays on transactions that touched Chase, not a failure of the whole Zelle network or of every bank involved. It's also worth noting this was an on-premises fault. As UC Today points out, the week's lesson isn't that cloud is fragile and the data center is safe, or the reverse.
The FAA: a router flood and a nationwide ground stop
The FAA incident was the most dramatic. In its official Q&A, the agency said its legacy equipment became unstable when data began flooding the FAA FTI network, at a rate of 1,000 connection attempts every 4 to 5 seconds. This disrupted the system and created instability with aircraft communications and radar feeds.
The FAA's initial investigation pointed to two potential locations as the reason for the excessive data flooding the systems: a radar site in Amarillo, Texas and a radio communications site in El Dorado, Arkansas. When these inconsistencies in the data happened, L3 Harris blocked those two locations and the network stabilized. Axios reported that L3Harris, the FAA's network service provider, has been deploying a new router at the Arkansas facility, and Technology.org reported that Duffy said L3Harris had replaced the Amarillo router three weeks earlier.
UC Today adds that Transportation Secretary Sean Duffy described crews replacing copper lines with fiber at El Dorado when the router started broadcasting across the network. Broadcasting has been disabled on all 69 comparable routers while the investigation continues, and the Arkansas and Texas units are being examined at an L3Harris lab in Florida with the FAA and FBI involved. Yahoo News reported Duffy saying an FAA team was "on site" at L3Harris's Florida lab to inspect the router that he said is the "likely cause" of Friday's nationwide ground stop.
Some context on scale:
- Technology.org reported that the ground stop began on Friday, November 13, and lasted nearly five hours.
- Air traffic controllers were credited for safely landing 3,500 planes that were already in the air when the nationwide ground stop was issued, according to Yahoo's live coverage.
- The incident marked the third nationwide ground stop, following a 90-minute outage in 2023 and on Sept. 11, 2001.
- The FAA said it and L3Harris continue to investigate the root cause.
UC Today also covers the theory that AI was to blame. The FAA's SMART AI-supported traffic-management system had expanded nationwide shortly before the incident. The FAA, the Department of Transportation and supplier Air Space Intelligence all said SMART wasn't involved, and FAA Administrator Bryan Bedford said investigators found no evidence linking the incident to AI.
For network engineers, this is a familiar failure: a broadcast storm from newly deployed hardware on a flat, legacy network. Disabling broadcast on every router of the same model is the quick fix. Whether segmentation, rate limiting or change-window policy should have contained it is a question for the root-cause investigation. One sign of how seriously that is being taken: the FAA's Q&A directly addresses why router replacement work was done in the middle of the day rather than at night.
Cisco links AI security findings to ServiceNow
The roundup's second section is about tooling. UC Today reports that Cisco announced AI security and observability features meant to give security and IT operations teams more context when an AI agent, configuration or network asset creates risk:
- Model Context Protocol and ServiceNow connectors added to Cisco's AI security portfolio. Cisco says they turn exposure findings into asset-linked problem records that go into existing remediation workflows, so teams don't have to manually reconcile vulnerability data, asset inventories and ownership details.
- Cisco IQ and Cisco Data Fabric pull threat, lifecycle and configuration data into Splunk for investigation.
- Evaluation models and Evaluation Tuner are generally available, according to Cisco.
- Tokenomics, a tool meant to connect AI spending with output quality, is scheduled for private preview.
The timing suits Cisco's pitch. Alerts don't resolve outages. Teams need to know which service is affected, which asset is involved and who owns the next step. These capabilities are Cisco's claims as reported by UC Today. Independent testing of the connectors hasn't been published.
IBM and Microsoft launch Imagine Enterprise
For Microsoft customers, the most strategic item is Imagine Enterprise, an IBM and Microsoft platform for building, running and governing AI agents for industry-specific processes. UC Today reports that it combines Microsoft products across AI, cloud, data, identity and security with more than 6,000 prebuilt AI agents and assets from IBM Consulting.
- The first industry frameworks cover financial services, insurance and retail. Manufacturing, energy, public sector and mobility are planned later.
- IBM says the platform connects agents to existing enterprise data and processes and includes controls, observability and AI-compliance support.
- It's available through IBM Consulting-led deployments, co-innovation engagements or self-managed options.
The two halves of the roundup are related. As businesses put agents on top of identity, data and network services, an outage like the Exchange Online incident doesn't just stop email. It can also stop every agent that depends on that mailbox, calendar or identity layer. UC Today asks what happens when the underlying platform, identity service, data source or network connection disappears. Neither the IBM–Microsoft announcement nor this week's incidents answer that.
Also noted
UC Today also reports that Zoom launched AI Launchpad, which offers eligible AI-native startups up to $150,000 in equity-free credits for its speech, language, voice and decision-intelligence APIs.
What Windows and Microsoft 365 admins should do this week
These incidents had different causes: a cloud network change, a DR environment interacting with production, an on-premises software bug and a misbehaving router. No evidence links them. PagerDuty CEO John DiLullo told UC Today that "Resilience is built before the outage, not during it." Here's what that means in practice:
- Map Exchange Online dependencies. List the workflows, approvals, Power Automate flows and agents that read or write mailbox and calendar data. EX1550543 showed that losing Exchange Online affects more than Outlook.
- Keep an out-of-band communication channel. If Teams, email and the Microsoft status page are all struggling, you need a phone tree or a separate messaging tool, along with a contact list that isn't stored only in the system that's down.
- Check DR isolation. Confirm that failover environments can't connect to production databases unintentionally. Workday's incident shows this can happen.
- Check integrations after recovery. When a SaaS tenant comes back, review failed scheduled jobs and integrations instead of assuming they caught up on their own.
- Restrict and review network changes. Microsoft's freeze on deployments and configuration changes and the FAA's question about midday maintenance both point to change control as a basic safeguard.
- Separate downstream delays from platform failures. The Chase and Zelle example applies to IT triage too. Before declaring a platform outage, find out whether it's actually down or only slow because something it depends on is.
Bottom line
None of the four outages involved a cyberattack, and the officials and companies involved ruled out AI as a cause in each case where that question came up. What they share is dependency and recovery risk. Meanwhile the industry, including Microsoft and IBM with Imagine Enterprise, is adding AI agents on top of those same infrastructure layers. Microsoft's post-incident report on EX1550543 and the FAA's root-cause findings will be the next documents worth reading.
References
- This Week: Outages, Agent Security and AI Platforms - UC Today UC Today · 2026-11-17T15:01:00.265000+00:00
- What we know about what caused the FAA's nationwide ground stop axios.com
- Router Install Behind Nationwide Ground Stop technology.org