Approve an AI system for a specific use
The decision
Should this model version be deployed under the proposed conditions in a liability-exposed enterprise process?
Accountable in this example
The designated approval body, with preparation by the business function, AI leads and Compliance.
What makes the decision difficult
Evaluation results, operating limits and human-oversight requirements sit in different records. It remains unclear whether they concern the same model version and intended use.
Before the decision
Epistemis makes missing evaluations and conflicting assumptions about the intended use visible. Accountable people can see which regulatory and internal requirements remain open and which alternatives were examined.
When the state is recorded
SEAL binds the model version, operating limits, evidence and accountable approval to the decision state.
An evaluation defined as necessary but still missing prevents the state from being recorded as final within Epistemis.
During a later review
Internal Audit or authorized external reviewers can use the Evidence Pack to examine:
Which evaluations covered this model version, which limitations were known and who approved its use?