When the threat model is "a fire takes out the DC", sure. But "the servers are root-level compromised", whoops, that's worth rebuilding from scratch. No way to make sure whatever AI agent did the compromise didn't go and embed malware in some flash ROM.