Loading…
Glossary term
Research and practice aimed at making AI systems reliable, controllable and free of serious harm.
AI safety is the research and engineering aimed at making AI systems reliable, controllable and not seriously harmful. It spans near-term issues such as misuse, security and bias, and longer-term ones about increasingly capable, autonomous systems.
Methods include evaluations, red teaming, interpretability and alignment techniques.
As AI is deployed widely, safety work shapes what is released and how it is regulated.