Security teams need a framework for incorporating protection throughout the AI lifecycle. Which approach best fits?

1. Measure subgroup performance and investigate representativeness and bias.
2. Use Google's Secure AI Framework as guidance alongside concrete controls.
3. Apply data minimization, approved processing, and least-privilege access.
4. Establish monitoring, evaluation ownership, and controlled update procedures.