SeterusnyaPanduan seterusnya
The EU General-Purpose AI Code of Practice
Teknikal
PANDUAN Teknikal
The EU AI Act places four main duties on providers of general-purpose AI models: technical documentation, downstream-provider information, a copyright-compliance policy, and a public summary of training content.
These are provider duties, with a narrow free/open-source exception for only some of them; they have applied since August 2, 2025, with a transition for certain earlier models.
Chapter V of the AI Act regulates providers of general-purpose AI models (GPAI). Under Article 53(1), providers must draw up and keep current technical documentation about the model, including training, testing, and evaluation; give downstream AI-system providers specified information about capabilities and limitations; put in place a policy to comply with EU copyright law and rights reservations; and publish a sufficiently detailed summary of training content using the AI Office template. These requirements support transparency and downstream compliance. They do not all require the same disclosure. Technical documentation is supplied to authorities upon request. Integration information goes to providers intending to incorporate the model in their systems, with protection for confidential business information and trade secrets. The training-content summary is public. The copyright duty is a policy obligation; it does not decide whether any particular training use infringes copyright or resolve ongoing legal questions. Article 53(2) gives a limited exception to the technical-documentation and downstream-information duties in points (a) and (b) for certain models under free and open-source licenses when parameters, architecture, and usage information are public. That exception does not cover the copyright policy or training summary, and it does not apply to systemic-risk GPAI models. Third-country providers must appoint an EU authorised representative under Article 54 before placing a GPAI model on the Union market, subject to Article 54(6)’s limited open-source exception. Providers of systemic-risk models also have Article 55 duties. The core GPAI obligations applied from August 2, 2025. The Commission says models placed on the market before that date have until August 2, 2027 to comply; Commission guidelines explain the Commission’s interpretation but are not legislation. Organizations should identify the provider for each model release, check whether significant modifications or other facts change that role, and keep evidence linked to the applicable model version. This guide is an overview, not legal advice.
Keputusan seni bina memacu prestasi dan kos operasi selama bertahun-tahun.
Pendidikan teknikal membantu pasukan memilih timbunan yang betul, bukan hanya yang terbaharu.
Pilihan kejuruteraan yang lebih baik mengurangkan insiden kebolehpercayaan dalam pengeluaran.
Commission templates, guidelines, and harmonised standards may evolve, and the Act allows delegated updates to documentation requirements. The August 2027 transition for some legacy models is a compliance deadline, not a general delay for new models. Providers should monitor EUR-Lex and the AI Office, preserve dated versions of documentation and summaries, and reassess obligations when a model is materially modified or designated systemic-risk. Preserve model-by-model version records and date the source of every training-summary template used. Recheck before release. Record who owns that review.
A model provider keeps technical information updated so an authority can request it, while protecting trade secrets under the Act’s rules.
A foundation-model supplier prepares integration information describing capabilities and limitations for downstream AI-system providers.
An open-source provider checks whether its license and public model information meet Article 53(2) before claiming the limited documentation exception.
A startup that fine-tunes and redistributes another model checks the Commission’s nonbinding guidance on when significant modifications may make it a provider.
Mengoptimumkan satu penanda aras boleh menyembunyikan kelemahan sistem yang lebih luas.
Kos infrastruktur dan penyelenggaraan sering dipandang remeh.
Jurang keselamatan dan pemerhatian boleh berkembang apabila sistem menjadi lebih kompleks.
Tentukan sasaran kependaman, kualiti dan kos sebelum pelaksanaan.
Penanda aras di bawah beban realistik dan keadaan data.
Pemantauan instrumen untuk ralat, drift dan kesan pengguna.
Sediakan laluan balik dan tindak balas insiden sebelum penskalaan.
Free newsletter
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
The EU AI Act places four main duties on providers of general-purpose AI models: technical documentation, downstream-provider information, a copyright-compliance policy, and a public summary of training content. These are provider duties, with a narrow free/open-source exception for only some of them; they have applied since August 2, 2025, with a transition for certain earlier models.
Article 53(1)(d) requires a public summary of training content according to the AI Office template.
Article 53(1)(b) requires documentation that helps downstream providers understand model capabilities and limitations and comply with the Act.
Article 53(1)(c) requires a policy to comply with Union copyright law and identify and comply with rights reservations.
Article 53(2) limits the exception to points (a) and (b), if the license and public-information conditions are met; it does not apply to systemic-risk models.
Article 53(1)(a) requires the provider to keep documentation available to authorities upon request.
Teruskan belajar
Lebih banyak panduan dipilih untuk topik ini
SeterusnyaPanduan seterusnya
The EU General-Purpose AI Code of Practice
Teknikal