In our 50-fault benchmark, Claude Code with a shell launched its own pods on 13 faults. On one, it dropped two MongoDB c...

In our 50-fault benchmark, Claude Code with a shell launched its own pods on 13 faults. On one, it dropped two MongoDB collections and still submitted the wrong diagnosis. The benchmark's prompt said a fix stage would follow, allowed resource changes, and told it not to ask for confirmation.The other arm had read-only tools only, and averaged 64 s to a diagnosis against 191 s.https://radarhq.io/blog/ai-sre-give-your-agent-context-first?utm_source=mastodon&utm_campaign=ai-sre-give-your-agent-context-first-launch-2026-10&utm_content=masto#Kubernetes #SRE #DevOps #AI