Security & Responsible AI
How we keep agents safe in production.
Operations are not the place to find out an AI agent “hallucinated.” Every agent we ship has a defined scope, a human approval gate where it belongs, an audit trail, and a way to turn it off in one click.
Human approval on high-risk actions
Customer-facing messages, financial transactions, and irreversible writes default to draft/approve mode. The agent prepares the action; a human signs off before it ships. Autonomous send is opt-in per workflow, never opt-out.
Audit logs on every agent action
Every trigger fire, decision, and output is logged with timestamp, inputs, and the resulting action. Owners and admins can replay any decision after the fact.
Override paths
Any agent can be paused, disabled, or rolled back instantly from the admin. Agents that depend on external systems gracefully degrade if a dependency is down — they do not fail-silent.
Approval gates
High-impact workflows require explicit sign-off at defined points — for example, a billing change suggestion, a refund, or a public review response. The agent surfaces the recommendation; a human accepts or revises.
No black-box automation for sensitive workflows
Every agent ships with a documented trigger, logic, output channel, measurement, and override path. If a workflow is sensitive, it must be inspectable.
System monitoring and error alerts
Agents are monitored in production. Errors, integration failures, and rate-limit hits raise alerts to a defined channel. Silent failure is treated as a defect, not an outcome.
Scoped data access
Each agent is given access to the systems and data it needs for its job — no more. Credentials are stored encrypted at rest (AES-256-GCM) and decrypted only at use. Least privilege is the default, not the aspiration.
Rollback and disable controls
Every deployed agent has a documented kill switch. Disabling an agent does not corrupt downstream state — workflows fall back to manual operation cleanly.
Data handling
- HTTPS everywhere, including service-to-service traffic.
- Session cookies are HMAC-signed and short-lived.
- Credentials at rest are AES-256-GCM encrypted with a key held outside the application.
- No PII or secrets in logs.
- 90-day minimum log retention; longer retention available per engagement.
- Subprocessors documented in the privacy policy and contractually bound to their own published terms.
Reporting a security concern
If you believe you have found a vulnerability or security issue, email security@operably.ai with a description and steps to reproduce. We will acknowledge within one business day.
Stop executing. Start governing.
The worst case: you do the mapping session and leave with a clearer picture of what's costing you — before spending anything on a build.
Start with an operations audit →