How AI behaves inside ChronoVault.

Commitments

Human review of every canonical change. Model outputs are attributed and auditable. Customers can opt out of individual features. No customer data is used to train shared models without explicit written consent. Rejection feedback is captured but not silently used to mutate models.

How it works in practice

When an AI feature runs, its output is recorded as a staged object with the model run identifier, the input it consumed, and the time it ran. A compliance manager reviews the staged object in a review queue and either promotes or rejects it. Only after promotion does the change affect canonical data.

Governance

Aggregate AI output quality is monitored by a dedicated governance function. Issues — hallucinations, systematic errors, prompt-injection attempts — are logged, investigated, and remediated on a documented cadence. Customers can request governance reports during procurement and renewal.

Limits we are honest about

AI in ChronoVault is good at reading structured regulator text, drafting requirements, suggesting mappings, and classifying evidence. It is not good at novel regulation interpretation, judgment calls, or negotiation. We do not pretend otherwise and we do not ship AI features that require the model to be something it is not.

Customer controls

Every AI feature has an opt-out. Customers can run ChronoVault with no AI features enabled if they prefer. Opt-outs are granular — customers can keep some AI features and disable others.