Glossary term
Glossary term
Safety and Alignment
A severe failure mode where humans cannot reliably predict, constrain, interrupt, or recover from AI system behavior. It is a frontier AI concern when autonomy, scale, speed, and tool access combine. This term should be reserved for severe scenarios; using it precisely helps senior leaders separate ordinary operational risk from frontier safety concerns.
The Statement on AI Risk (May 2023) signed by Geoffrey Hinton, Yoshua Bengio, and Sam Altman names loss of control among extinction-level risks.
The Center for AI Safety's research agenda explicitly includes loss-of-control scenarios as a priority for technical safety work.
Anthropic's RSP and OpenAI's Preparedness Framework both explicitly reference loss-of-control concerns as a justification for pause-and-evaluate protocols.