← sources

Microsoft had to clarify their outage was “not related to AI / agents.” Made me rethink how much access I give Claude Code.

source: reddit (reddit:ClaudeCode)published 2026-11-13 18:50ingested 2026-11-14 02:41

retained text

Microsoft’s statement on the recent M365 outage explicitly said it was “a technical issue not related to cyber or AI / agents.” It was a network change, nothing to do with agents.

But the fact that they *had* to say it stuck with me. People’s first assumption now is that an agent did something it shouldn’t have.

That made me look at my own Claude Code setup more honestly:

* How many tools and commands have I auto-approved just to stop the prompts?
* Does it have access to anything with production credentials?
* If it ran something destructive in a session I wasn’t watching, would I even know?

For solo projects it’s probably fine. But I work with companies rolling out Copilot and agents across whole organisations, and “what’s the blast radius if an agent misbehaves?” is now a real question from leadership.

Curious how people here handle it:

1. Do you keep permissions tight, or go fully permissive and rely on git to undo things?
2. Anyone using hooks or deny rules to block specific commands?
3. Any setup you’d recommend for teams, not just solo devs?

document fetched 2026-11-14 00:44 · canonical https://reddit.com/r/ClaudeCode/comments/1xvg7ps/microsoft_had_to_clarify_their_outage_was_not