Security & governance

Every workflow we build follows a strict approval-first, least-privilege architecture. Here is exactly how we handle security, data access, and AI governance.

🛡️ Core principles

Human approval before every production action

No agent writes to production systems, sends external communications, or modifies live data without explicit human approval. Every sensitive action goes through a staging-review-approve cycle.

Least-privilege access

Each workflow agent gets the minimum permissions required to do its job. Read-only by default. Write access is scoped to specific resources and revocable at any time.

Sandbox before production

All new workflows run in a sandboxed environment first. Changes are tested in staging with sample data before any production deployment. Filesystem and network access is isolated.

Full audit trail

Every agent action is logged — what was read, what was generated, what was approved, by whom, and when. Audit logs are retained and exportable for compliance purposes.

🏗️ Architecture layers

Our delivery architecture has four control layers, from outermost (most restricted) to innermost.

Human approval
Every production-facing action requires explicit human sign-off. Approvers are named workflow owners inside your organization.
Permission controls
API-first integrations with scoped tokens. MCP connections to trusted servers only. No credentials stored in agent prompts.
Staging environment
All changes preview in staging before touching production. Test cases validate output quality and format before approval.
Agent execution
Claude Code runs in a sandboxed environment with filesystem and network isolation. Read-only by default. Every tool call is permission-gated.

🔒 Data handling

⚖️ AI Act readiness

The EU AI Act (Article 4) requires organizations to ensure AI literacy for people working with AI systems. Our engagements include governance elements by default.

We do not build high-risk AI systems

We explicitly avoid HR decision-making, credit/insurance scoring, healthcare diagnosis, safety-critical systems, or any use case classified as high-risk under the EU AI Act. Our focus is low-risk operational automation: document processing, reporting, content generation, and internal tooling — where the human stays in the loop and final decisions are made by people.

Delivery checklist

Every pilot and retainer engagement follows this security checklist.

Questions about security?

We are happy to walk through our architecture and governance approach in a discovery call.

Book a discovery call →