A team plans to address AI security once the system is working. Why does the lifecycle framing argue against that?

• ABecause security controls cannot be applied to a running system.
• BBecause training data gathered without access control, or a pipeline nobody can audit, cannot be secured retrospectively without redoing the work.
• CBecause model performance degrades if controls are added later.
• DBecause the largest exposure is at deployment, which happens first.