Chapter V · Section 3 · Obligations of providers of general-purpose AI models with systemic risk
Article 55 — Obligations of providers of general-purpose AI models with systemic risk
▼ Primary text, verbatim. Our annotations appear below, visibly separated.
1. In addition to the obligations listed in Articles 53 and 54, providers of general-purpose AI models with systemic risk shall:
(a) perform model evaluation in accordance with standardised protocols and tools reflecting the state of the art, including conducting and documenting adversarial testing of the model with a view to identifying and mitigating systemic risks;
(b) assess and mitigate possible systemic risks at Union level, including their sources, that may stem from the development, the placing on the market, or the use of general-purpose AI models with systemic risk;
(c) keep track of, document, and report, without undue delay, to the AI Office and, as appropriate, to national competent authorities, relevant information about serious incidents and possible corrective measures to address them;
(d) ensure an adequate level of cybersecurity protection for the general-purpose AI model with systemic risk and the physical infrastructure of the model.
2. Providers of general-purpose AI models with systemic risk may rely on codes of practice within the meaning of Article 56 to demonstrate compliance with the obligations set out in paragraph 1 of this Article, until a harmonised standard is published. Compliance with European harmonised standards grants providers the presumption of conformity to the extent that those standards cover those obligations. Providers of general-purpose AI models with systemic risks who do not adhere to an approved code of practice or do not comply with a European harmonised standard shall demonstrate alternative adequate means of compliance for assessment by the Commission.
3. Any information or documentation obtained pursuant to this Article, including trade secrets, shall be treated in accordance with the confidentiality obligations set out in Article 78.
This text is meant purely as a documentation tool and has no legal effect. The Union's institutions do not assume any liability for its contents. The authentic versions of the relevant acts, including their preambles, are those published in the Official Journal of the European Union and available in EUR-Lex.
Recital 110 — interpretive context
General-purpose AI models could pose systemic risks which include, but are not limited to, any actual or reasonably foreseeable negative effects in relation to major accidents, disruptions of critical sectors and serious consequences to public health and safety; any actual or reasonably foreseeable negative effects on democratic processes, public and economic security; the dissemination of illegal, false, or…
Recital 114 — interpretive context
The providers of general-purpose AI models presenting systemic risks should be subject, in addition to the obligations provided for providers of general-purpose AI models, to obligations aimed at identifying and mitigating those risks and ensuring an adequate level of cybersecurity protection, regardless of whether it is provided as a standalone model or embedded in an AI system or a product. To achieve those…
Recital 115 — interpretive context
Providers of general-purpose AI models with systemic risks should assess and mitigate possible systemic risks. If, despite efforts to identify and prevent risks related to a general-purpose AI model that may present systemic risks, the development or use of the model causes a serious incident, the general-purpose AI model provider should without undue delay keep track of the incident and report any relevant…
What this means for you
In your terms · Systemic-risk model obligations
Adversarial testing of the model is conducted and documented; the runs and their records are engineering work.
- Documented adversarial test runs for the model
In your terms · Systemic-risk model obligations
Evaluations, incident tracking and model-weight security are operational programs with schedules and evidence, not one-off exercises.
- Evaluation program with adversarial testing
- Model-weight security controls
Failure smells likeThe evaluation that would have found the capability was a benchmark run once, and nothing adversarial has been pointed at the model since.
In your terms · Systemic-risk model obligations
If your training compute approaches the presumption threshold, the notification and obligations clock starts with the training run, not the launch.
- Training-compute accounting
Obligations derived from this article
Scenarios that touch this provision
An open model, fine-tuned and sold inside your product
Illustrative Reading still settling
You took an open-weight model, fine-tuned it on your own data, and it now powers a paid feature in the product you sell.
- Your role
- Provider
- Where it lands
- Not classified In force 2 Aug 2025
- Decided by
- Chapter V, as the provider of the modified model, and separately whatever Article 6 says about the system you built on it.
What applies
What you have to be able to produce
- Downstream information pack
- Model documentation
- Onboarding notes for AI-touching roles
- Provider-status assessment for fine-tunes
- Public training-content summary
- Team enablement plan for people operating AI systems
- Written mandate for the model's authorised representative
What would change the answer
- The Article 53(2) exception covers models released under a free and open-source licence with public parameters, architecture and usage information, and it never covers a model with systemic risk.
- Recital 103 reads components provided against a price or otherwise monetised as outside the free and open-source exceptions. A recital is interpretive context, not operative law, and it is the weakest link in this row.
- Crossing the Article 51 systemic-risk threshold adds the Article 55 duties and removes the exception entirely.
If you would rather not read the law
The basics page explains the Regulation's own categories in order: scope, role, tier, date. The engineering view groups the obligations by the platform capability they demand.