general-purpose AI model with systemic risk
In plain wordsThe 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.
Under the EU AI ActEU AI Act, prop. n.: The European Union's law on artificial intelligence, the first broad AI law anywhere: it bans some uses, sets strict checks for risky ones and puts duties on the makers of the biggest AI models., a general-purpose AI modelgeneral-purpose AI model, n.: The EU's legal name for the large, adaptable AI models behind chatbots and similar tools: models that can turn their hand to many different jobs and be built into many other products. bearing the heaviest duties because its capabilities rank with or above those of the most advanced models (high-impact capabilities) or because the Commission finds its capabilities or impact equivalent, weighing Annex XIII indicators such as parameters, data, training compute, modalities, benchmark results and user reach. Training with more than 10²⁵ floating-point operations (FLOPFLOP, n.: A single arithmetic step, such as multiplying or adding two numbers. Counting them measures how much computing went into an AI model, and laws use such counts to decide which models face extra rules.) creates a presumption of high-impact capabilities; the provider must tell the Commission within two weeks of crossing, or expecting to cross, that line and may argue that its model poses no systemic risk. The extra duties are to assess and reduce systemic risks across the EU, backed by model evaluation to current standards, including adversarial testingred teaming, n.: Deliberately attacking your own AI system — trying to make it misbehave, leak secrets or help with harm — so that weaknesses are found before real adversaries or users find them.; to keep the model and its infrastructure adequately secure; and to track and promptly report serious incidentsserious incident, n.: Under EU AI law, an AI failure that kills or badly injures someone, disrupts vital services, violates fundamental rights or seriously damages property — which the maker must report to authorities quickly. to the AI OfficeAI Office, prop. n.: The European Commission's team that oversees the EU AI Act, especially the rules for the largest AI models, and coordinates how the law is applied across Europe.. 'Systemic risk' means risk peculiar to such capabilities that could spread through the value chain and seriously affect public health, safety, security, fundamental rights or society in the EU.
Crossing 10²⁵ FLOP obliged the lab to notify the AI Office that its model had systemic risk.
Origin
Introduced in the political agreement of December 2023 as a compromise between the Parliament, which wanted binding rules for all foundation models, and France, Germany and Italy, which favored self-regulation through codes of conduct. The 10²⁵ FLOP figure, roughly the training compute of GPT-4-class models of 2023, can be revised by delegated act.
Usage note
Providers usually show compliance through the Safety and Security chapter of the General-Purpose AI Code of PracticeGeneral-Purpose AI Code of Practice, prop. n.: A voluntary rulebook, published by the EU in 2025, that makers of large AI models can sign to show they follow the EU AI Act's rules on transparency, copyright and safety.. Critics argue that a fixed 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. will capture more models as compute gets cheaper while missing efficient but capable ones; the Commission can designate models below it.
Related
- general-purpose AI modelgeneral-purpose AI model, n.: The EU's legal name for the large, adaptable AI models behind chatbots and similar tools: models that can turn their hand to many different jobs and be built into many other products.
- frontier AIfrontier AI, n.: The most capable general-purpose AI systems of the moment — those at or beyond the leading edge — which governments single out for extra safety attention.
- 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 Code of PracticeGeneral-Purpose AI Code of Practice, prop. n.: A voluntary rulebook, published by the EU in 2025, that makers of large AI models can sign to show they follow the EU AI Act's rules on transparency, copyright and safety.
- AI OfficeAI Office, prop. n.: The European Commission's team that oversees the EU AI Act, especially the rules for the largest AI models, and coordinates how the law is applied across Europe.
- dangerous capability evaluationdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself.
- serious incidentserious incident, n.: Under EU AI law, an AI failure that kills or badly injures someone, disrupts vital services, violates fundamental rights or seriously damages property — which the maker must report to authorities quickly.
Defined in law or standards
EU AI Act (Reg. (EU) 2024/1689), Art. 3(64)–(65), Arts. 51–52 and Annex XIII; obligations Art. 55
Mentioned in
EU AI ActEU AI Act, prop. n.: The European Union's law on artificial intelligence, the first broad AI law anywhere: it bans some uses, sets strict checks for risky ones and puts duties on the makers of the biggest AI models., frontier AIfrontier AI, n.: The most capable general-purpose AI systems of the moment — those at or beyond the leading edge — which governments single out for extra safety attention., 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., general-purpose AI modelgeneral-purpose AI model, n.: The EU's legal name for the large, adaptable AI models behind chatbots and similar tools: models that can turn their hand to many different jobs and be built into many other products., 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 AIgeneral-purpose AI, n.: AI that can do a wide range of different tasks, like writing, coding, translating and analyzing images, rather than the one job it was built for. Today's chatbots are the main example., General-Purpose AI Code of PracticeGeneral-Purpose AI Code of Practice, prop. n.: A voluntary rulebook, published by the EU in 2025, that makers of large AI models can sign to show they follow the EU AI Act's rules on transparency, copyright and safety., serious incidentserious incident, n.: Under EU AI law, an AI failure that kills or badly injures someone, disrupts vital services, violates fundamental rights or seriously damages property — which the maker must report to authorities quickly., dual-use foundation modeldual-use foundation model, n.: A 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.