Policies
Cortex checks before every execution which approvals and policies exist and maintains audit-proof records.
Set your rules
Require your approval
Draw your boundaries
Keep your proof
Protect your trade secrets
A contractual promise does not protect data at the point where it reaches a model. Cortex enforces data protection as a rule of the platform. At the system boundary, before a model is called, per data class, proven in every run.
Rule, record, result
An automation without a rule before it is a risk, one without a record is a blind spot. Cortex gives every execution a rule before it, a record during it and a result after it, the human approves where the rule requires it. Every step remains traceable as an event.
At the system boundary
A promise that data will not leave the company is worth as much as the point at which it is checked. Cortex enforces the rule per data class at the system boundary before a model is called, anything impermissible is removed or replaced with consistent substitutes, the rest stays in the company. Which model the company chooses changes nothing about this.
Replaced, not deleted
Deleted data leaves gaps that distort an answer. Cortex detects personal and third-party content before storage and replaces it with consistent substitute content, and if a model reconstructs what was removed in an answer, this is detected. The log names class and count, never the content.
Proof per run
Audit and certification do not ask whether a system meant well, but what it did. Cortex records source, rule, author, timestamp and result for every run, and every alert remains traceable with source, cause and recipient. The trail is in place before anyone asks for it.
Visible omissions
An answer that conceals omissions appears complete and is not. Cortex states in every answer what was omitted due to the policies and why. The person who decides knows the limits of their answer.