// Security & trust
Trust is earned before it's asked for.
We built AutoOps the opposite way from most privileged software — it proves its behavior to you first, and every capability that could change your systems is off until you deliberately turn it on.
Three independent layers of control
Is it even enabled?
Restart, patch, block — every active capability is off by default. You opt in, per capability, per tenant. Say nothing and AutoOps only ever watches.
Human, or autonomous?
Even when enabled, the default is supervised — a person confirms every action. Hands-off mode is a separate, deliberate choice, never an accident.
Emergency stop
One kill-switch halts every active behavior instantly, fleet-wide, the moment you want it stopped. Break-glass, always within reach.
Privilege tiers
Access is layered and explicit. Each tier is a separate, disclosed grant — nothing escalates silently.
Read-only
Reads system metrics and health signals. The default. Cannot change anything.
Log access
Reads security-relevant logs (e.g. auth) to detect intrusions. Still read-only.
Blocking
May execute an approved block command — a separate, disclosed grant, off unless you turn it on.
Your data stays yours
Metadata, not content. Briefly kept. Yours to delete.
We read health, not your data
System metrics and security log signals — never your files, databases, or application data.
No secrets, no credentials
The agent isn't built to collect keys or credentials. It reports what happens to the machine, not what's inside it.
Encrypted & isolated
Encrypted in transit, strictly multi-tenant. Your data is scoped to you and never visible to another client.
Kept briefly, deleted on demand
Telemetry auto-purges on a rolling window. One-click exit removes the agent and your data. No lock-in.
And you can verify every word of this
The Silent Observer trial doesn't just prove we take no actions — it lets your own team watch exactly what the agent does and reports before you commit to anything. Trust, but verify. We're counting on it.