Safety Alignment
Noun · AI & Machine Learning
Definitions
The work of shaping a model so its behavior stays within intended safety, policy, and usefulness boundaries in realistic use.
In plain English: Training and tuning AI so it behaves in safer, intended ways.
Example: "Safety alignment reduced harmful completions but made some benign tasks more cautious."