Which term describes an approach to safely developing AI systems by encoding guiding principles?

• A. Alignment Science Blog
• B. RLHF (Reinforcement Learning from Human Feedback)
• C. Constitutional AI (CAI)
• D. Frontier Red Team Blog

Answer: (C)

Encoding guiding principles as a constitutional framework means designing a set of rules or norms that the AI must follow when reasoning and generating outputs. This approach, often called Constitutional AI, treats a fixed “constitution” of principles—values, constraints, and priorities—that guide the model’s decisions and help ensure safer, more predictable behavior. The model can be evaluated against these principles, justify its outputs by showing alignment with the rules, and defer or revise if a response would violate the constitution. This creates an interpretable safety mechanism based on codified guidelines rather than relying solely on data or human feedback.

By contrast, reinforcement learning from human feedback relies on humans to rate or correct outputs and shape behavior through reward signals, which is a different alignment pathway that doesn’t hinge on encoding a fixed set of constitutional rules. The other items are not established methods for encoding guiding principles into an AI’s behavior.