EU adopts rules to mitigate societal risks from next-generation AI

The new regulations require GPAI vendors, which are models capable of generating language and trained with large amounts of computation (over 10²³ FLOPs), to clearly disclose training data and strictly adhere to copyright rules.

This is a pioneering step by the EU in building a comprehensive legal framework, aiming at developing AI in a responsible, transparent and safe manner.

The European Commission (EC) has published detailed guidance to clarify the obligations of each stakeholder. A standard template has also been published for model developers to use in summarizing training data, helping regulators assess the transparency and risk of each AI product.

In addition, a voluntary Code of Practice developed by independent experts has also been recognised by the EC as being in line with the requirements of the Act. AI providers that sign up to and comply with this Code will enjoy reduced regulatory burdens.

From 2 August, all new GPAI models introduced to the EU market will have to comply with the new rules. For models already in existence before then, the deadline to update and ensure compliance is 2 August 2027.

For AI models that are assessed as advanced or pose a high systemic risk (training beyond 10²⁵ FLOPs), the EU requires additional measures such as prior notification to the EC, the development of security mechanisms, post-audit monitoring, and technical safety assurance.

The EU's official adoption of regulations on GPAI is expected to create a model for ethical technology development, while protecting intellectual property rights and minimizing social risks from new-generation AI.

Source: https://baoquocte.vn/eu-ap-dung-quy-dinh-giam-thieu-rui-ro-xa-hoi-tu-ai-the-he-moi-323075.html