I gave an AI SSH access to my production infrastructure – 3 months later
18 hours ago
- #Regulated Environments
- #IT Infrastructure
- #AI Copilot
- Using Claude AI as a co-pilot for on-prem IT infrastructure management in a regulated Swiss financial firm, including SSH access to devices for configuration reading, auditing, and documentation.
- Key achievements: complete switch security audits uncovering critical findings (e.g., active HTTP servers, missing ACLs, unpatched CVEs), structured documentation from real configs in a Git repo, storage audits reducing findings by two-thirds, and migration to an open-source monitoring stack (Prometheus, Grafana, Loki, Wazuh).
- Limitations highlighted: AI cannot make business decisions, replace human experience in context, manage human relationships, and requires human oversight due to potential errors; all access is read-only to mitigate risks.
- Practical guidance for IT managers: start with read-only access, focus on documentation first, then proceed to audits, deploy monitoring, and always maintain control with validation and Git versioning.
- Tools used include Claude Code for SSH connectivity, self-hosted Git (Gitea), and open-source monitoring solutions, emphasizing a shift from reactive to proactive operations with AI as a force multiplier for small teams.