EU introduces new rules for general-purpose AI models

At stake are Generally Accessible Artificial Intelligence Models, that is, systems developed with a focus on accessibility, transparency and widespread use, that is, AI that can be used by many, unlike closed or restricted models.

They are now subject to new rules on documentation, transparency, security, copyright compliance and risk assessment in the EU.

Examples include models trained in or capable of generating language, such as OpenAI's ChatGPT, Google Gemini, Anthropic's Claude, and Mistral's generative models. These are considered general-purpose models because they can perform diverse tasks, such as writing text, translating, answering questions, generating images, assisting with programming, and more.

Therefore, starting this Saturday, new suppliers selling AI models covered by this regime must comply with the new rules, and those operating systemic risk models will have to notify the office created by the European Commission for this purpose and mitigate such risks.

Existing models on the market will have until August 2, 2027 to adapt.

In the case of ChatgGPT, as it already exists on the market, the company will have another two years to fully adapt. For now, the EU strongly recommends that suppliers like OpenAI adopt the current code of conduct, serving as preparation for full compliance by 2027.

In August 2024, the EU's AI law came into force, the first global legislation for this technology. It aims to safeguard fundamental rights within the EU, but within which only some provisions are currently applicable.

The regulation establishes obligations for AI based on its potential risks and level of impact, stipulating the prohibition of certain applications, limitations on the use of biometric identification systems by law enforcement authorities, exemptions applicable to law enforcement, obligations for high-risk systems and transparency requirements.

It also provides for fines for violations.