Frontier AI
Highly capable, general-purpose AI models at the leading edge of development, whose risks and capabilities are not yet fully understood.
Frontier AI↗ refers to highly capable, general-purpose AI models operating at the very leading edge of what current technology can do, systems whose full range of capabilities, risks, and failure modes are not yet completely understood, even by the organizations that build them. The term is used specifically to distinguish these cutting-edge systems from smaller, narrower, well-understood AI tools.
Because frontier models can sometimes exhibit unexpected or emergent capabilities that weren't explicitly designed in, researchers and policymakers have increasingly focused on frontier AI as an area needing extra safety testing, red-teaming, and oversight before wide deployment. Governance frameworks in several countries now specifically single out frontier-scale models for additional scrutiny given their scale and the uncertainty around their full range of behavior.