Frontier AI model
General-purpose models near cutting‑edge performance.
How the laws define this
Both instruments establish that frontier AI models represent the forefront or leading edge of AI capabilities, characterized by broad applicability across a wide range of tasks. Where they diverge is in their primary regulatory focus. The National Security Memorandum on Artificial Intelligence (NSM on AI, Oct 24, 2024) defines general-purpose AI models approaching or achieving this performance in terms of the elevated risk they pose, subjecting them to heightened governance, testing, and security measures within the AI Framework. Conversely, Winning the Race: America's AI Action Plan explicitly addresses advanced large language models and other cutting-edge systems that often require significant computational resources. Instead of focusing on testing or security frameworks, this plan emphasizes ensuring these systems protect free speech and American values, and are free from ideological bias.
Synthesised from the 2 statutory definitions below. Each one is quoted in full, with its source.
What it triggers
- Under the National Security Memorandum on Artificial Intelligence, it is subject to heightened governance, testing, and security measures within the AI Framework.source
- The Winning the Race: America's AI Action Plan emphasizes ensuring these frontier models protect free speech and American values, and are free from ideological bias.source
Definitions (2)
A general-purpose AI model that approaches or achieves leading-edge performance across a wide range of tasks, posing elevated risk and therefore subject to heightened governance, testing, and security measures within the AI Framework.
Advanced large language models (LLMs) and other cutting-edge AI systems that represent the forefront of AI capabilities, often requiring significant computational resources and possessing broad applicability. The plan emphasizes ensuring these frontier models protect free speech and American values, and are free from ideological bias.
Not to be confused with
Frontier model refers specifically to the most capable AI models needing enhanced oversight, while frontier AI model denotes general-purpose models near cutting-edge performance.
Foundation models/frontier AI are highly capable, general-purpose models with broad applicability, aligning closely but distinguished by specific emphases in each definition.
High-risk AI focuses on uses with potential for severe or large-scale harm, whereas frontier AI model emphasizes model capabilities and performance.
Questions and answers
What obligations apply to frontier AI models under US policy?
The Winning the Race: America's AI Action Plan recommends ensuring they protect free speech and American values while being free from ideological bias. The National Security Memorandum requires heightened governance, testing, and security measures.
How is a frontier AI model defined in the National Security Memorandum?
It is a general-purpose AI model approaching or achieving leading-edge performance across tasks, posing elevated risk.
What makes the US AI Action Plan's definition of frontier AI unique?
It highlights significant computational resources, broad applicability, and the need to protect free speech and American values without ideological bias.
Related Terms
Frontier Model
Most capable AI models requiring enhanced oversight....
Foundation models / Frontier AI
Highly capable, general-purpose AI models with broad applicability....
Frontier models
Very large or general-purpose models with cross-sector impact....
Open-weight model
A model whose weights are publicly available....
High‑risk AI
AI uses with potential for severe or large‑scale harm....