General-purpose AI model (GPAI)

A model designed to perform a broad range of tasks and generate outputs across multiple contexts rather than being limited to a specific narrowly defined purpose; the Act specifies special transparency, documentation and governance obligations for GPAI, including requirements for provider disclosures and risk mitigation when outputs are used in the Union.

A model designed or intended to perform a broad spectrum of tasks rather than being tailored to a specific application; GPAI are subject to Chapter V obligations including systemic risk identification and mitigation, tailored technical documentation, transparency duties and cooperation with the European Artificial Intelligence Office.

AI models designed and marketed for general-purpose use across diverse tasks and contexts rather than a specific narrowly defined purpose; the Act imposes specific documentation, testing, watermarking/metadata and mitigation measures on GPAI providers to address scale and systemic risks, and empowers the Commission to set further obligations and fines for non-compliance.

AI models designed and deployed to perform a wide range of tasks across different contexts and domains rather than a narrowly defined specific purpose; the Act defines GPAI and subjects their providers to specific transparency, documentation, safety assessment and Commission oversight obligations, including methodologies for identifying systemic risk. GPAI may attract additional regulatory scrutiny due to potential systemic effects and broad deployability.

An AI model not designed for a single narrow purpose but usable for numerous downstream systems; the Digital Omnibus proposal centralises oversight of many systems built on certain GPAI categories, requiring clear delimitation of types triggering AI Office competence.

GPAI models are defined as AI models displaying significant generality and capable of competently performing a wide range of distinct tasks, often trained with large amounts of data using self-supervision at scale.