What does “Safety Case” mean?

Model output constrained to a defined schema such as JSON, tool arguments, or typed fields.

A structured way for one software system to send requests to and receive responses from another system.

The degree to which an AI response is supported by source data or retrieved evidence.

A moderation layer that blocks or rewrites unsafe model inputs or outputs.

A high-priority instruction that sets behavior, policy, and response style for a model.

A hypothetical AI system that can perform most intellectual tasks at a human level across many domains.