ProductOverviewHow It WorksCapabilitiesFor Your Team
TrustOverviewData Protection & GovernanceSecurity & AccessOwnership & ConfidentialityAudit & Accountability
ResourcesTechnical OverviewAPI OverviewSupport
CompanyAbout AscentMLContactRed Mountain Innovations
Talk to AscentML

Know what changed, who changed it, and when

Machine learning decisions should be as traceable as any other business decision.

AscentML records important actions across your environment and keeps every prediction connected to the model and data that produced it.

What is recorded

Nearly every state-changing action is audited, including:

People and access

  • Sign-ins and failed sign-in attempts
  • User invitations and role changes
  • API key creation, rotation, and revocation
  • Privileged administrative actions

Data and models

  • Dataset uploads, releases, and deletions
  • Training requests and status changes
  • Model publication
  • Deployment and traffic allocation changes
  • Quota and entitlement changes

What each record contains

An audit record is written to answer an investigation, not just to exist.

  • Timestamp
  • Customer and user
  • Action and affected entity
  • Before and after state
  • Request ID and source

Records are append-only and redacted at the time they are written, so sensitive values are not stored in the audit trail.

Audit detail — before-and-after state for a deployment change.

One request, one thread to follow

Every request is assigned a correlation ID.

That same ID follows the request through the API, background jobs, internal processing, inference, and the audit trail. When something needs investigating, your team and ours are looking at the same thread.

Error responses include the request ID, so a support conversation can start with exactly what happened.

Every prediction has a history

Datasets and model versions are immutable, and deployments are stored as versioned revisions. That makes it possible to trace any result back to its source:

The prediction that was produced → the deployment that served it → the model version that made it → the data that model was built from.

Per-prediction explanations can be generated later against the exact model version that made the original prediction.

Visible to the people who need it

Audit history is not buried in a settings page. Projects, datasets, models, and deployments each include their own audit view, available to users with the appropriate permissions.

Records can be filtered, searched, and exported, with before-and-after differences shown for each change. Notifications can alert the right people when important events occur, such as a deployment change or an exceeded quota.

Have audit or review obligations of your own?

Tell us what your organization needs to demonstrate. We will show you what AscentML records and how your team can access it.