How does AI Control detect changes in AI communication?

It compares current communication against the language baseline and flags protocol changes (for example JSON to UNKNOWN), new tokens, new sequences and new capability associations. When high-impact capabilities such as a production API are involved, the change is marked REVIEW REQUIRED for policy review. A change is never an automatic block or authorization.

Frequently asked questions

Is a detected change a security incident?

Not automatically. It is a signal for review that may raise risk depending on capability and context.