Frontier AI
Highly capable, cutting‑edge AI models (a subset of general‑purpose/foundation models) whose scale or capabilities may create systemic or catastrophic risks.
How the laws define this
Regulatory sources consistently identify frontier AI as advanced, highly capable general-purpose models or systems whose capabilities match, exceed, or go materially beyond state-of-the-art or widely deployed systems, presenting novel or systemic risks that demand specialized safety testing, empirical evaluation, and oversight. However, these instruments diverge on the severity and domains of harm included within the concept. The Bletchley Declaration and the AI Security Institute explicitly extend the term to models posing catastrophic risks, with the Bletchley Declaration specifically highlighting impacts across cybersecurity, biotechnology, and information integrity. Conversely, the AI Safety Institute and the AI Opportunities Action Plan frame the risk threshold around novel or systemic risks without explicitly citing catastrophic outcomes. Furthermore, while the AI Opportunities Action Plan highlights potential future regulation as a consequence, the AI Security Institute uniquely emphasizes that the term is an operational and currently contested label.
Synthesised from the 4 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first. Showing 3 of the 4 instruments that define this term; all are listed under Definitions below.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| United Kingdom | AI Safety Institute (establishment following AI Safety Summit) | In Force | Emphasizes novel or systemic risks for specialised empirical evaluation and monitoring by the AI Safety Institute, without specifying model capabilities relative to existing systems. |
| United Kingdom | AI Opportunities Action Plan (Matt Clifford) and Government acceptance/response | In Force | Highlights models materially beyond widely deployed systems, treating them as needing specific safety testing, oversight, and potential future regulation. |
| United Kingdom | AI Security Institute (renaming / rebrand of AI Safety Institute) | In Force | Operationalizes as highly capable general-purpose models matching or exceeding advanced systems, presented as a contested label for special evaluation and oversight. |
What it triggers
- Under UK AI Safety Institute policy, frontier AI systems are prioritised for focused pre- and post-deployment testing and international coordination.source
- Under UK AI Opportunities Action Plan, frontier AI models are described as those whose capabilities are materially beyond widely deployed systems and which may create novel or systemic risks.source
- Under UK AI Security Institute policy, frontier AI models are subject to evaluation of national security risks, voluntary testing and cross-government coordination.source
- Under the Bletchley Declaration, countries commit to cooperate on safe development of frontier AI through shared scientific understanding.source
Definitions (4)
Advanced AI systems whose capabilities may present novel or systemic risks requiring specialised empirical evaluation and monitoring. The Institute uses this term to prioritise systems for focused pre- and post-deployment testing and international coordination.
Described as highly capable general-purpose AI models (including foundation models) and other systems that exhibit advanced capabilities that may match or exceed those present in state-of-the-art models, and which could pose systemic or catastrophic risks across domains such as cybersecurity, biotechnology, and information integrity.
Models whose capabilities are materially beyond those of widely deployed systems and which may create novel or systemic risks; the Plan treats these as requiring specific safety testing, oversight and potential future regulation.
Highly capable general‑purpose models whose capabilities match or exceed the most advanced systems in use and which could present new systemic or catastrophic risks; the term is presented as an operational (and currently contested) label for models requiring special evaluation and oversight.
Not to be confused with
General-Purpose AI Model refers to any scalable model competent across tasks, while frontier AI specifically denotes those with advanced capabilities posing systemic or catastrophic risks.
General-Purpose AI Model with Systemic Risk is an EU legal designation triggering heightened obligations, unlike frontier AI which is a UK policy term without binding duties.
Questions and answers
How does the AI Opportunities Action Plan address frontier AI?
It describes frontier AI models, which are materially beyond widely deployed systems, as those that may create novel or systemic risks.
Related Terms
General-Purpose AI Model
An AI model trained at scale that displays significant generality and can competently perform many distinct tasks and be integrated into varied downstream systems....
Foundation Model
A large AI model trained on broad data that is designed to be adapted for many different tasks (e.g., LLMs, multimodal models)....
Dual-Use Foundation Model
Large foundation model with potential civilian and national security applications....
General-Purpose AI Model with Systemic Risk
A general-purpose AI model legally designated as carrying risks that can affect the EU market or society at scale, subject to heightened obligations under the EU AI Act....