EU’s AI Act Enters New Enforcement Phase

(EQRoy/Shutterstock)

The European Union’s AI Act hit another stage of enforcement over the weekend that will impact how companies can build and implement AI systems that impact or are used by European residents.

It’s been one year since the EU’s sweeping new AI Act went into effect. The first stage of enforcement of the law–which was first proposed by the European Commission in 2021 and was formally approved in March 2024–began in February 2025. It banned certain uses of AI deemed to have unacceptable risks, such as gathering of facial images from the Internet or closed-circuit TV.

On August 2, the second phase of enforcement began, which introduces two main requirements. The first is the requirement that all EU member states have established their national authorities to provide notification and surveillance. The second requirement is the beginning of enforcement of regulations concerning so-called “general purpose” AI (GPAI) models, which include language models and computer vision systems.

The AI Act has several requirements for GPAI models, including disclosure of training data and usage licenses. This means GPAI model providers must provide a detailed summary of the content used to train the model as well as proof of consent of people who generated the training data. For GPAI models gauged to pose “systemic risk,” the model providers must show how they’re evaluating the models, how they’re mitigating the risk, and report on any serious incidents.

“The sources used to train a general-purpose AI model that is made available to users in Europe wil have to be clearly documented,” Thomas Regnier, a European Commission spokesperson, told France24. “If they are protected by copyright, the authors will have to be remunerated and, above all, their consent will have to be obtained.”

The pyramid of criticality used by the AI Act

Enforcement is set to begin for any new GPAI models put into production after August 2, 2025. For GPAI models that are already in production–such as those from US tech giants Google, OpenAI, Meta, and Anthropic as well as European AI company Mistral–the European Commission is providing another year’s grace period before enforcement begins. Failures to adhere to the new law can incur fines ranging range from €7.5 million (about $8.1 million) or 1% of turnover all the way up to €35 million (about $38 million) or 7% of global revenue. These fines are now in force.

To encourage compliance, the European Commission last month published the EU AI Code of Practice, a statement that’s intended to help companies comply with AI Act obligations on safety, transparency, and copyright issues. Many US tech giants and European AI firms signed it, while others did not. Google signed it, but not without reservations.

“…[W]e remain concerned that the AI Act and Code risk slowing Europe’s development and deployment of AI,” the company wrote in a blog post. “In particular, departures from EU copyright law, steps that slow approvals, or requirements that expose trade secrets could chill European model development and deployment, harming Europe’s competitiveness.”

Facebook parent Meta said it won’t sign the AI Code of Practice. “Europe is heading down the wrong path on AI,” Joel Kaplan, the head of Meta’s global affairs office, wrote on LinkedIn. “We have carefully reviewed the European Commission’s Code of Practice for general-purpose AI (GPAI) models and Meta won’t be signing it.”

The next phase of enforcement will involve so-called high risk AI systems, which the European Commission describes as AI systems used in areas like law enforcement, education, critical infrastructure, and credit scoring. Organizations deploying these types of systems will need to take extra precautions before they can deploy them, such as carrying out a risk assessment to determine if they violate fundamental rights, provide monitoring of systems, maintaining logs of these AI systems’ activities, and ensuring that support staff are trained.

Related Items:

Five Questions as the EU AI Act Goes Into Effect

EU Votes AI Act Into Law, with Enforcement Starting By End of 2024

European Policymakers Approve Rules for AI Act