Security & Trust

Reliability isn't
bolted on.

The number-one objection from businesses is: "what if it does something wrong?" Our entire method is designed around answering exactly that.

Our principles

Five things every AI Employee inherits.

Safety isn't a feature we layer on top — it's how the workers are built from the start.

Review-mode-to-autonomy ladder

Every Digital FTE starts with humans approving every output. Autonomy is earned by proving reliable, not granted by default.

10-80-10 safeguard

A human owns the spec at the start. The worker executes in the middle. A human verifies the output at the end. Always.

Hard guardrails & escalation

Clear, spec-defined limits the worker physically cannot exceed. Uncertain cases route to a human automatically.

Complete audit logs

Every tool call, every output, every decision — recorded, timestamped, and exportable for review or compliance.

Versioned & reversible

Every change to a worker's spec is version-controlled. Roll back to a known-good state instantly if something goes wrong.

Written data agreements

Scoped access, defined retention, clear DPA. You always know what data the worker can see and how long it's kept.

Data handling

You choose where your data lives.

Three honest deployment options — pick what matches your data-control needs.

Default

Cloud

Managed by us with scoped access and signed data agreements. Most capable, easiest to update.

Premium

Client-cloud / VPC

Runs inside your own cloud account (AWS, GCP, Azure). Data never leaves your perimeter.

Enterprise

On-prem / local

Runs on your hardware for strict residency needs. Uses smaller local models.

Operational controls

Built for the security team's checklist.

Scoped access

Workers see only what they need. Tool-level permissions, not global access.

Encryption in transit & at rest

Industry-standard TLS for transport, AES-256 for storage. Keys rotated regularly.

Real-time monitoring

Live observability on every worker. Anomalies surface within seconds.

Incident response

Documented runbooks. Defined escalation paths. Post-incident reports shared.

Defined data retention

Logs and conversation history kept for a configurable window. Then deleted.

SSO & access controls

Single sign-on, role-based permissions, and review workflows on the client dashboard.

DPA & MSA on request

We'll sign a Data Processing Agreement and customize the Master Services Agreement.

Compliance roadmap

SOC 2 and GDPR-style controls on our roadmap. Happy to share progress on call.

Responsible AI

Honest about what AI can — and can't — do.

A model with a clever prompt is not an employee. We don't pretend otherwise. Every Digital FTE we ship comes with explicit boundaries, escalation paths, and humans-in-the-loop. We'd rather over-escalate than over-confidence.

If a worker isn't ready for autonomy, it stays in review mode. If a request falls outside its spec, it routes to a human. If something goes wrong, the audit log shows exactly what happened.

Workers operate within spec-defined limits — never beyond
Uncertainty routes to humans by default, not the other way around
No training on your data without explicit written consent
Model providers selected based on capability + privacy posture
Portable across providers — no lock-in if you switch later
Security review

Need a vendor security questionnaire?

We're happy to walk through your security process, share our controls documentation, and answer your team's questions on a call.