AI Alignment

Level 1

Making sure the AI does what we actually want and doesn't do anything bad — like teaching a pet to follow the rules.

Level 2

The effort to make sure AI systems behave the way humans intend, following our values and goals instead of doing something unexpected or harmful.

Level 3

The research field focused on ensuring AI systems act in accordance with human intentions, values, and ethical principles, especially as systems become more capable.

Level 4

The technical and philosophical challenge of specifying, encoding, and verifying that an AI system's objectives and behaviors remain consistent with human values and intentions across diverse contexts.

Level 5

The superalignment problem: ensuring that arbitrarily capable optimization processes remain corrigible and value-aligned — encompassing inner alignment (mesa-optimizer objectives match training objectives) and outer alignment (training objectives capture human intent).

Sources

• NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.