# AI Alignment

The field of ensuring that an AI system's goals and behavior stay consistent with human values and intentions.

**AI alignment[↗](/ai-alignment)** is the field of research and engineering focused on ensuring that an AI system's actual behavior and goals stay consistent with the values and intentions of the humans who built and deployed it. An unaligned system might technically achieve the goal it was given while doing so in ways its creators never intended or wanted.

Alignment becomes harder as AI systems become more capable and are given more autonomy to act in the world, since it's increasingly difficult to anticipate every situation the system might encounter and specify exactly the right behavior in advance. Techniques like reinforcement learning from human feedback, careful testing, and red-teaming are all part of the broader effort to keep AI behavior aligned with human intent as the technology continues to advance.