Glossary · Evaluation & safety
Alignment
The effort to make a model or AI system behave in ways that match intended goals, constraints, and human preferences across both expected and adversarial situations.
Why it matters
A system can optimize the stated metric while violating the user's real intent, so alignment requires evaluation, oversight, and system controls as well as model training.
Related terms
Browse the learning paths to see this term in context — every lesson is free to read.