GPAI Code of Practice: systemic risk obligations under Article 55
As of 18 November 2026 · checked against Regulation (EU) 2024/1689 and the Commission Code of Practice page
Information only, not legal advice. This page summarises the Code of Practice and the obligations in Articles 51 to 55. Confirm any compliance decision with qualified counsel.
Articles 51 to 55 of Regulation (EU) 2024/1689 split general-purpose AI models into two tiers: GPAI, and GPAI with systemic risk. The second tier carries materially heavier obligations. The General-Purpose AI Code of Practice is a voluntary tool, endorsed by the Commission, with three chapters: Transparency, Copyright, and Safety and Security. Signatories can rely on it to demonstrate compliance (Art. 53(4) and 55(2)). This page covers what separates the tiers, what Article 55 adds, and how the operational mechanics (notification, model evaluation, incident reporting) actually work.
GPAI vs. GPAI with systemic risk
Every general-purpose AI model is in scope of Article 53. A subset crosses into Article 55. Two paths into the systemic-risk tier:
- Compute threshold (Art. 51(2)). Cumulative compute used for training exceeds 1025 floating-point operations. The Commission can adjust this threshold by delegated act.
- Designation (Art. 51(1)(b)). The Commission, acting on a qualified alert from the scientific panel or its own assessment, designates the model as having capabilities of high impact. Criteria are listed in Annex XIII: number of parameters, dataset size, modalities, downstream reach, registered users, and others.
Once the threshold is met, Article 52(1) requires the provider to notify the Commission without delay and in any event within two weeks of meeting the threshold or, if later, of being aware that it will be met.
The obligations in Article 53 (every GPAI)
- Maintain technical documentation of the model, kept up to date (Art. 53(1)(a), Annex XI).
- Make information and documentation available to downstream providers integrating the model (Art. 53(1)(b), Annex XII).
- Put in place a policy to comply with Union copyright law, including respect for Article 4(3) DSM Directive opt-outs (Art. 53(1)(c)).
- Draw up and publish a sufficiently detailed summary of the training content (Art. 53(1)(d)).
- Cooperate with the Commission and competent authorities (Art. 53(3)).
What Article 55 adds (systemic-risk tier)
- Model evaluation (Art. 55(1)(a)). Perform model evaluation in accordance with standardised protocols and tools reflecting the state of the art, including adversarial testing, with a view to identifying and mitigating systemic risks.
- Risk assessment and mitigation (Art. 55(1)(b)). Assess and mitigate possible systemic risks at Union level, including their sources, that may stem from development, placing on the market, or use of the model.
- Serious incident tracking and reporting (Art. 55(1)(c)). Keep track of, document, and report without undue delay to the AI Office and, where appropriate, to national competent authorities, relevant information about serious incidents and possible corrective measures.
- Cybersecurity protection (Art. 55(1)(d)). Ensure an adequate level of cybersecurity protection for the model and its physical infrastructure.
The Safety and Security chapter of the Code translates these four duties into commitments that signatories accept. Adherence is voluntary. Providers who do not adhere must demonstrate alternative adequate means of compliance for assessment by the Commission (Art. 55(2)).
The 14-day notification
Article 52(1) requires notification within two weeks of the model meeting the systemic-risk threshold. The notification:
- Goes to the European Commission.
- Must include the information necessary to demonstrate that the requirement has been met.
- Allows the provider to argue, with sufficiently substantiated reasons, that despite meeting the criteria the model exceptionally does not present systemic risks (Art. 52(2)).
- If the Commission finds the arguments not sufficiently substantiated, it rejects them and the model counts as having systemic risk (Art. 52(3)). The Commission can also designate a model of its own motion (Art. 52(4)). The Regulation sets no decision deadline for the Commission.
The Commission publishes and keeps up to date a list of GPAI models with systemic risk (Art. 52(6)).
Model evaluation in practice
Article 55(1)(a) requires evaluation with standardised protocols and tools reflecting the state of the art, including documented adversarial testing. How a signatory plans, runs and repeats evaluations is set out in the Safety and Security chapter of the Code. Read it in the original on the Commission page; this page does not paraphrase its commitments.
Serious incident reporting
The reporting obligation under Art. 55(1)(c) parallels Article 73 for high-risk AI systems but applies to the GPAI provider directly, towards the AI Office. Incident reports themselves are not public. AI Act Radar emits events on the gpai topic when the Commission publishes something about them, such as guidance or enforcement action. Illustrative payload:
{
"id": "evt_example_gpai_guidance",
"type": "guidance",
"source": "aioffice",
"topics": ["gpai"],
"severity": "medium",
"title_en": "AI Office publishes guidance on serious incident reporting for GPAI models",
"ai_act_articles": ["Art. 55(1)(c)"],
"url_official": "https://digital-strategy.ec.europa.eu/..."
} Cybersecurity expectations
Article 55(1)(d) requires an adequate level of cybersecurity protection for the model and its physical infrastructure. Recital 115 names what that protection should consider: accidental model leakage, unauthorised releases, circumvention of safety measures, cyberattacks, unauthorised access and model theft. It suggests securing model weights, algorithms, servers and data sets through operational security measures, cybersecurity policies, established technical solutions, and cyber and physical access controls.
What to track
- Code of Practice updates and clarifications from the AI Office.
- Commission delegated acts adjusting the 1025 FLOP threshold or Annex XIII criteria.
- New entries in the public register of GPAI models with systemic risk.
- Scientific panel alerts that may trigger designation under Art. 51(1)(b).
Sources
- Regulation (EU) 2024/1689, Articles 51 to 55 and recital 115: EUR-Lex
- Commission, The General-Purpose AI Code of Practice: digital-strategy.ec.europa.eu
→ See live updates: https://aiactradar.com
→ Subscribe to webhook: https://docs.aiactradar.com/quickstart