Nearly 24 hours without Outlook, Teams or Copilot, across more than 70 countries. Microsoft says the cause was a network upgrade, not AI — and that is exactly the point I want to press on. The AI layer inherits every crack in the plumbing beneath it. Here is what I would ask of Microsoft, of Indian CIOs, and of the regulators now asking questions.
On 13 November, Microsoft posted the sentence that matters: "This was a technical issue not related to cyber or AI / agents. The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change" (BBC News live coverage, 13 November). It said it was "deeply sorry" and promised a full post-incident report.
Reuters reported that day that disruption ran past 24 hours, with no timeline for full recovery. Downdetector's count, as Reuters carried it:
- reports from 74 countries within about an hour;
- 7.4 million reports in total;
- a peak near 20,000 a minute, roughly 1,500 times normal.
Anderson Economic Group, a Michigan consultancy, put possible lost productivity at $7 billion; Microsoft has published no figure of its own (BetaNews, 13 November).
Now the paragraph the AI industry should read twice. Yahoo Finance reported on 13 November that Copilot could not reach users' email, files or calendars — and that Microsoft's connectors for OpenAI's ChatGPT and Anthropic's Claude broke with it. The AI did not fail. The floor gave way beneath it. "Not AI" is not a defence; it is the diagnosis. An assistant is only as reliable as the mailbox it is permitted to read.
AI strategist Walter Pasquarelli told the BBC that as assistants are plugged into our mail, calendars and files, "a failure like this may spread even further."
Nor was it one company's bad day. The same Friday, every departing flight in the United States was grounded after a separate communications failure — only the second such nationwide alert since 9/11 — and the same digest noted Indian businesses, from startups to multinationals, feeling this outage in internal and customer-facing work (The Hindu, 14 November).
I have written this prescription before, for smaller systems. In Answer Sheet Portal Chaos (20 May 2026) I asked the CBSE for a "manual fallback process" and a "clear public timeline". My ResumesExchange white paper (26 April 2026) put the principle in one line: "Central database = single point of failure ... P2P network = cannot be shut down." AI only multiplies how much work halts when the centre halts.
If I were advising a CIO in Mumbai or a regulator in Washington this week, I would ask for five things:
- A declared degraded mode: a cached, read-only fallback that works without the AI at all.
- A status page hosted off your own cloud, so it is reachable when the cloud is not.
- A named manual path for month-end — phone tree, paper register, and the fax machine Reuters reported a Taiwanese firm using to confirm trades.
- Automatic service credits. Microsoft's 99.9% promise and 25–100% credits mean little if the customer's IT department must file the claim (Yahoo Finance, 13 November).
- The promised post-incident report, published in full, in the spirit of an aviation accident report.
I believe the AI stack will be judged less by how clever it is than by what it does when the layer beneath it breaks. For nearly a day this week, we found out. Washington is asking questions. In India, we should be too.
With regards,
Hemen Parekh
www.Iam-Immortal.ai / www.IndiaAGI.ai / www.YourContentCreator.in
14 November 2026
No comments:
Post a Comment