概述
It has transparency, copyright, and systemic-risk safety and security chapters; providers may instead show compliance through other adequate means or harmonised standards.
深入探討
Article 56 of the EU AI Act encourages the AI Office to facilitate codes of practice that support proper application of the Regulation. The General-Purpose AI Code of Practice was published on July 10, 2025, after independent experts and stakeholders developed its text. It has three chapters: Transparency and Copyright for GPAI providers generally, and Safety and Security for providers of models with systemic risk. The Commission’s current page lists signatories and states that the code is an adequate voluntary tool. The code translates duties into practical documentation and processes. Its Transparency chapter supports information providers must keep or give downstream developers, while the Copyright chapter supports the required policy for complying with EU copyright law and rights reservations. The Safety and Security chapter offers practices for systemic-risk evaluation, assessment, mitigation, incident handling, and cybersecurity. It is not a general AI-system code and does not replace Article 53 or Article 55. Signing is voluntary. Under Articles 53(4) and 55(2), providers may rely on an approved code to demonstrate compliance until harmonised standards are published. A provider that does not adhere to an approved code or applicable standard must demonstrate alternative adequate means for Commission assessment. Following the code is therefore a route to show compliance, not a product certificate, blanket authorization, or immunity from enforcement. Regulation (EU) 2026/1744 amended Article 56 to require the Commission and the AI Board to monitor codes and assess whether they adequately cover the GPAI obligations, with public assessment. The current Commission page reports its status through July 31, 2026. Providers should verify the latest approved version, signature status, and standards when planning compliance. This is an informational overview, not legal advice.
現實世界的實施
A provider signs the Transparency and Copyright chapters and uses the model documentation form to organize Article 53 evidence.
A systemic-risk provider compares its safety framework and incident process with the Safety and Security chapter before deciding whether to sign.
A provider that does not sign builds an alternative-compliance record because the Act still applies.
A compliance lead checks Commission updates on code adequacy, signatories, and harmonised standards rather than treating a past signature as permanent approval.
The Future of The EU General-Purpose AI Code of Practice
The Code of Practice may be updated as standards and implementation evidence develop. Regulation (EU) 2026/1744 makes adequacy review and ongoing monitoring explicit, so provider reliance should track the Commission’s current assessment. Providers can prepare an alternative compliance case if they do not sign or if a chapter fails to cover a specific duty. Check the current AI Office page before each major compliance review. Monitor any assessment published after the Commission’s latest review. Check the current Code before renewal.
付諸實踐
為什麼這很重要
- 成本與預算. 多年來,架構決策決定著效能和營運成本。
- 更明確的決策. 技術教育幫助團隊選擇正確的堆疊,而不僅僅是最新的堆疊。
- 品質管控. 更好的工程選擇可以減少生產中的可靠性事故。
留意
- 優化一項基準測試可以隱藏更廣泛的系統弱點。
- 基礎設施和維護成本常常被低估。
- 隨著系統變得更加複雜,安全性和可觀察性差距可能會擴大。
後續步驟
- 在實施之前定義延遲、品質和成本目標。
- 在實際負載和資料條件下進行基準測試。
- 儀器監控錯誤、漂移和使用者影響。
- 在擴展之前準備回滾和事件回應路徑。
不斷探索
常見問題
What is The EU General-Purpose AI Code of Practice?
The EU General-Purpose AI Code of Practice is a voluntary tool published in July 2025 to help GPAI providers demonstrate how they meet AI Act obligations. It has transparency, copyright, and systemic-risk safety and security chapters; providers may instead show compliance through other adequate means or harmonised standards.
何時發布了通用人工智慧行為準則?
The Commission states that the final GPAI Code of Practice was published on July 10, 2025.
通用人工智慧行為準則包含哪些章節?
The Code has Transparency and Copyright chapters and a Safety and Security chapter for systemic-risk models.
安全與保安章節適用於誰?
The Commission says the Safety and Security chapter concerns the additional obligations for systemic-risk GPAI providers.
哪一項說明了簽署通用人工智慧行為準則的法律地位?
The code is voluntary; providers may rely on it for covered obligations or demonstrate alternative adequate means.
未遵循已批准的代碼或協調標準的提供者需要展示什麼?
Articles 53(4) and 55(2) require providers using another route to demonstrate alternative adequate means for assessment.