Control

Control covers the settings and mechanisms used to change AI system behaviour and reverse a change. Changes should be deliberate, centrally managed, and recorded.

Run covers availability. Control covers changes and recovery. An organization may operate reliable workloads yet still need application code changes to move between model providers. Control addresses that gap.

Control in operation

Observe identifies a problem. Control provides the settings, access paths, and rollback mechanisms needed to correct it.

Control includes these capabilities:

• Change a prompt, a model, or a policy, and roll it back.
• Put every internal AI application behind shared ingress with SSO.
• Carry a person's identity all the way to the data.
• Say how long AI data lives, and prove it was deleted.
• Move a workload between models or providers without touching code.
• Let people work remotely without the data going with them.

Operational checks

• Time the rollback of a prompt change from last month. The process should not depend on finding the person who made it.
• Move one application to a different model provider without changing application code.
• Stop one internal AI application from reaching users in under five minutes without finding its owner.
• Identify which external person accessed a resource on a given day and confirm that the access ended with the engagement.

Remote access belongs in Control because data movement and user access must be managed together. Organizations that keep data inside a perimeter while supporting remote staff and consultants need workspaces that remain near the data. The same controls also govern cross-border access. See the questions.

Control and delivery speed

Control can slow Build when every change requires a queue or central review. Teams are more likely to use controls included in a fast, reusable delivery path. Manual queues encourage workarounds that leave fewer records.

Secure uses the records from these mechanisms as evidence for security and compliance review.