dual-use foundation model
In plain wordsA US legal term for a general-purpose AI model that serves everyday uses but is capable enough that it could help someone build weapons, mount cyberattacks or evade human control.
In US policy, a large general-purpose foundation modelfoundation model, n.: A very large AI model, built once at great cost from vast amounts of text, images or code gathered mostly from the internet, that many products then reuse as their starting point, from chatbots to coding tools. whose abilities could endanger national security, the economy or public health and safety. The criteria set by Executive Order 14110Executive Order 14110, prop. n.: A sweeping 2023 order by US President Joe Biden that told federal agencies how to manage AI risks and required makers of the largest AI models to report their safety tests to the government. It was revoked in 2025. were technical as well as functional: training on broad data, usually by self-supervision; tens of billions of parameters or more; usefulness across many settings; and strong performance, actual or easily obtained by modification, at dangerous work such as helping non-experts make chemical, biological, radiological or nuclear weaponsCBRN risk, n.: The danger that AI could help someone make or use chemical, biological, radiological or nuclear weapons — for example, by explaining hard technical steps an attacker could not work out alone., automating serious offensive cyber operations, or slipping out of human oversight by deception. Safeguards that users could get around did not take a model outside the definition. The term framed the order's reporting duties and a 2024 Commerce Department review of open-weightopen-weight model, n.: An AI model whose trained internal numbers are published for anyone to download and run on their own computers, even if the data and code used to make it stay private. releases, which recommended monitoring rather than restricting widely available weights. 'Dual-use' borrows the export-control sense of goods with both civilian and military applications.
The reporting requirement applied only to dual-use foundation models above the compute threshold.
Origin
Defined in Executive Order 14110 (October 30, 2023), combining the foundation modelfoundation model, n.: A very large AI model, built once at great cost from vast amounts of text, images or code gathered mostly from the internet, that many products then reuse as their starting point, from chatbots to coding tools. coinage of Stanford researchers (2021) with the long-standing export-control notion of dual-use items. The National Telecommunications and Information Administration's July 2024 report 'Dual-Use Foundation Models with Widely Available Model Weights' extended its use.
Usage note
Its legal force lapsed with the order's revocation in January 2025, though the term persists in policy writing and some bills. Not equivalent to the EU's general-purpose AI model with systemic riskgeneral-purpose AI model with systemic risk, n.: The EU's label for the very largest and most capable AI models, whose failure or misuse could cause harm across society, so their makers must test them for dangers and report serious incidents., which turns on 'high-impact capabilities', presumed above 10²⁵ FLOP of training compute, or on designation by the European Commission.
Related
- Executive Order 14110Executive Order 14110, prop. n.: A sweeping 2023 order by US President Joe Biden that told federal agencies how to manage AI risks and required makers of the largest AI models to report their safety tests to the government. It was revoked in 2025.
- foundation modelfoundation model, n.: A very large AI model, built once at great cost from vast amounts of text, images or code gathered mostly from the internet, that many products then reuse as their starting point, from chatbots to coding tools.
- frontier modelfrontier model, n.: One of the most capable AI models in existence at a given time, built by a handful of leading labs at great cost, and the main target of AI safety rules.
- open-weight modelopen-weight model, n.: An AI model whose trained internal numbers are published for anyone to download and run on their own computers, even if the data and code used to make it stay private.
- CBRN riskCBRN risk, n.: The danger that AI could help someone make or use chemical, biological, radiological or nuclear weapons — for example, by explaining hard technical steps an attacker could not work out alone.
- compute thresholdcompute threshold, n.: A legal cutoff based on how much computation was used to train an AI model. Models trained with more than that amount face extra rules, on the theory that bigger training runs produce more capable models.
- general-purpose AI model with systemic riskgeneral-purpose AI model with systemic risk, n.: The EU's label for the very largest and most capable AI models, whose failure or misuse could cause harm across society, so their makers must test them for dangers and report serious incidents.
Sources
- National Telecommunications and Information Administration (2024). Dual-Use Foundation Models with Widely Available Model Weights. US Department of Commerce.
Defined in law or standards
Executive Order 14110 (2023), Sec. 3(k) (revoked January 2025)
Mentioned in
open-weight modelopen-weight model, n.: An AI model whose trained internal numbers are published for anyone to download and run on their own computers, even if the data and code used to make it stay private., compute thresholdcompute threshold, n.: A legal cutoff based on how much computation was used to train an AI model. Models trained with more than that amount face extra rules, on the theory that bigger training runs produce more capable models., dual-usedual-use, adj.: Describes technology or knowledge that can be used both to help and to harm: the same AI that finds software bugs for defenders can find them for attackers., Executive Order 14110Executive Order 14110, prop. n.: A sweeping 2023 order by US President Joe Biden that told federal agencies how to manage AI risks and required makers of the largest AI models to report their safety tests to the government. It was revoked in 2025., marginal riskmarginal risk, n.: How much extra danger an AI model adds compared with what people could already do without it — for instance with search engines, textbooks or models that are already freely available.