Security
Predictable behavior, explicit access, honest posture.
Security is not a badge page. It's the design choices that decide what an agent can and cannot do, and how you find out about it.
Access model
Scoped, revocable, explained.
Every tool connection carries the exact scopes the agent uses, in language you'd write yourself.
Per-action scopes
Each agent action declares the connection and scope it needs. The agent cannot use a scope it did not declare.
Least privilege by default
Scopes are proposed to match the actions in the plan. Unused scopes are never requested.
Revocable in one click
Removing a connection stops in-flight actions and blocks new ones immediately.
Owner-only sensitive actions
High-impact actions (large refunds, mass changes) route to a named human owner - never auto-run.
Data protection
Encrypted in transit and at rest.
Data your agent handles lives in your tools. What Fact6 stores is limited to what the agent needs to run and to explain what it did.
TLS 1.2+ everywhere
All API traffic is served over TLS 1.2 or higher, HSTS enabled for the product surface.
Encryption at rest
Persistent data (configuration, logs, credentials) is encrypted at rest with per-tenant keys.
Credential handling
Tool credentials are stored encrypted and never returned in plaintext through the product.
Log retention
Activity logs default to 90 days. Configurable retention on Growth and Scale plans.
Approval guarantees
Nothing runs outside the rules you set.
Approval boundaries are enforced by the runtime - not by convention, not by prompt.
Runtime enforcement
Threshold and category rules are checked before every write action - if the check fails, the action does not happen.
Explicit escalation
Cases outside auto-run rules are routed to a named owner with the full context, not silently dropped.
Block-list of last resort
Certain classes of actions can be marked never-auto - the agent must always request a human.
Auditable rules
Approval rules are versioned. You can trace which version was in force at any point in time.
Compliance posture
Stated honestly, updated as it changes.
We only claim what is currently in place. Anything in progress is labeled as such.
Data protection principles
Purpose-limited processing, data minimization, and access controls scoped to declared uses.
Sub-processors
A current list of infrastructure and email sub-processors is available on request.
Incident response
Security incidents affecting customer data are communicated to affected customers within 72 hours of confirmation.
Formal certifications
Formal certifications are in progress. Anything not on this page has not been achieved yet.