Artificial intelligence is advancing at an extraordinary pace, and few technological developments have attracted as much public attention in recent months as the emergence of ChatGPT. The conversational AI system developed by OpenAI has demonstrated capabilities that, until very recently, seemed confined to science fiction, generating human-like text, answering complex questions, writing computer code, summarising documents, and producing original content within seconds. As governments, businesses, universities, and citizens begin to explore the opportunities offered by these new technologies, policymakers are also confronting an equally important challenge: how should artificial intelligence be regulated to encourage innovation while protecting fundamental rights and ensuring public trust?
Within the European Union, this debate is now entering a decisive phase through the development of the Artificial Intelligence Act (AI Act), a landmark legislative proposal that seeks to establish the world’s first comprehensive legal framework dedicated specifically to artificial intelligence. Rather than attempting to slow technological progress, the proposed regulation aims to create clear rules that allow innovation to flourish while ensuring that increasingly powerful AI systems remain transparent, accountable, and aligned with European values.
One of the principal figures behind this legislative initiative is Romanian Member of the European Parliament Dragoș Tudorache, who serves as the Parliament’s rapporteur for the AI Act. Speaking during a conference in Strasbourg, Tudorache explained how the proposed legislation would apply to generative AI systems such as ChatGPT if the regulation were already in force.
Greater Transparency for Generative Artificial Intelligence
According to Tudorache, developers of advanced AI models such as ChatGPT would be required to comply with a series of transparency obligations designed to provide greater clarity about how these systems are built and how they operate.
If the AI Act were already applicable, developers would need to provide detailed information about the datasets used to train their models, explain the general functioning of the system, and disclose its fundamental technical characteristics. These measures are intended to ensure that organisations using AI—particularly in applications that may later fall within higher-risk categories—have sufficient information to comply with European legal requirements.
As Tudorache explained:
“If our rules entered into force today, ChatGPT’s developers would have to consider the interests of downstream actors in the value chain, ensuring they provide the information necessary for compliance whenever the technology is deployed in high-risk contexts. They would also need to be transparent about the data used to train ChatGPT, explain how the system functions, and disclose its core parameters.”
One of the central ambitions of the AI Act is to reduce the opacity surrounding advanced AI systems, replacing the traditional “black box” approach with greater accountability and openness throughout the AI development process.
Copyright Becomes a Central Issue in AI Development
Among the most significant elements of the proposed legislation is its treatment of copyrighted material used during the training of generative AI models.
Large Language Models such as ChatGPT require enormous volumes of digital information to learn language, reasoning, and communication patterns. These training datasets may include books, newspapers, academic publications, websites, software repositories, and many other forms of copyrighted material.
Recognising the growing concerns expressed by authors, publishers, artists, and content creators, the European Parliament’s proposal introduces new transparency requirements obliging developers of general-purpose AI systems to document and disclose summaries of the copyrighted materials used during model training.
Tudorache also questioned whether sufficient consideration had originally been given to copyright issues during the development of ChatGPT, suggesting that future AI developers should be able to demonstrate that the legality of both their training data and the content generated by their systems has been taken into account from the earliest stages of development.
“Developers should demonstrate that they considered the legality of the content their systems would generate during the design of the algorithm. They must also disclose all copyrighted materials used to train their models and document them comprehensively.”
This proposal reflects a broader European objective: to improve transparency surrounding AI development while balancing technological innovation with the protection of intellectual property rights.
Will Regulation Prevent Europe from Building Its Own ChatGPT?
As negotiations over the AI Act continue, one of the recurring questions is whether stricter regulation could make it more difficult for European companies to compete with large technology firms from the United States.
Tudorache rejected this argument, maintaining that Europe’s greatest obstacle is not regulation but investment.
Developing frontier AI models comparable to ChatGPT requires extraordinary computational resources, highly specialised engineering expertise, and financial investments measured in hundreds of millions—or even billions—of euros. According to Tudorache, access to these resources represents a much greater challenge for European companies than regulatory compliance.
“There is certainly room for a European ChatGPT. The legislation is not what limits European companies. Developing a model like ChatGPT requires enormous resources, and access to those resources is a much greater challenge than these regulations.”
This perspective highlights one of the broader debates currently taking place within Europe. While policymakers seek to establish clear safeguards for artificial intelligence, many also recognise that Europe must significantly strengthen its research ecosystem, technological infrastructure, and investment capacity if it wishes to compete globally in advanced AI development.
The European Parliament Advances the AI Act
The European Parliament has now adopted Tudorache’s report on the proposed AI Act, allowing the legislation to move into the next phase of negotiations between the European Parliament, the Council of the European Union, and the European Commission.
These interinstitutional negotiations will determine the final shape of what is expected to become the world’s first comprehensive legal framework dedicated specifically to artificial intelligence.
Because of its ambitious scope, the AI Act is already attracting worldwide attention, with governments, technology companies, researchers, and international organisations closely following its development. Many observers believe that, just as the General Data Protection Regulation (GDPR) influenced privacy legislation around the world, the AI Act could establish a new international benchmark for AI governance.
A Risk-Based Regulatory Framework
Rather than applying identical rules to every AI application, the proposed legislation adopts a risk-based approach, meaning that regulatory obligations increase according to the potential risks presented by an AI system.
Applications considered to present an unacceptable level of risk would be prohibited altogether. These include AI systems designed to perform social scoring, classifying individuals according to their behaviour or personal characteristics in ways considered incompatible with European values and fundamental rights.
Members of the European Parliament have also expanded the list of prohibited practices to include several particularly intrusive applications of artificial intelligence, including:
- Real-time and post-event remote biometric identification in publicly accessible spaces.
- Biometric categorisation systems based on sensitive characteristics such as race, ethnicity, religion, gender, political beliefs, or citizenship.
- Predictive policing systems relying on profiling, location, or previous criminal behaviour.
- Emotion recognition technologies used in workplaces, schools, border management, or law enforcement.
- The indiscriminate collection of facial images from the internet or CCTV recordings for the creation of facial recognition databases.
Together, these proposed prohibitions illustrate the European Parliament’s intention to ensure that artificial intelligence develops in a manner that respects privacy, human dignity, democratic values, and fundamental rights.
Expanding the Definition of High-Risk AI Systems
The Parliament has also broadened the definition of high-risk AI systems, extending it beyond traditional safety-critical applications.
Under the current proposal, AI systems capable of significantly affecting health, safety, fundamental rights, democratic processes, or the environment would fall within this category.
Among the newly designated high-risk applications are:
- AI systems capable of influencing voters or election outcomes.
- Recommendation algorithms used by very large online platforms serving more than 45 million users.
Given the potential influence these technologies may have on democratic participation and public discourse, legislators believe they require particularly robust oversight and accountability mechanisms.
New Obligations for General-Purpose Artificial Intelligence
One of the most innovative aspects of the proposed AI Act is the introduction of a dedicated regulatory framework for general-purpose artificial intelligence models, a rapidly evolving area of AI development represented by systems such as ChatGPT.
Developers of these foundation models would be expected to:
- Assess and mitigate potential risks affecting health, safety, fundamental rights, democracy, the environment, and the rule of law.
- Register their models in a European Union database before making them available within the European market.
- Clearly disclose AI-generated content, helping distinguish authentic material from AI-generated content, including deepfakes.
- Implement safeguards against the generation of illegal content.
- Publish detailed summaries of the copyrighted data used during model training.
These measures reflect the growing recognition that general-purpose AI systems require governance mechanisms that extend beyond traditional sector-specific regulation.
Supporting Innovation While Protecting Citizens’ Rights
Although the AI Act introduces significant new obligations, the European Parliament also seeks to ensure that regulation does not unnecessarily hinder innovation.
To support research and technological development, Members of Parliament have proposed exemptions for scientific research activities and for certain open-source AI components. The legislation also promotes the creation of regulatory sandboxes, controlled environments established by public authorities where developers can safely test innovative AI systems before deploying them under real-world conditions.
At the same time, the proposal strengthens citizens’ rights by allowing individuals to submit complaints regarding AI systems and to receive explanations about decisions made by high-risk AI systems whenever those decisions significantly affect their fundamental rights.
Finally, the proposed legislation also strengthens the future role of the European Union AI Office, which will be responsible for supervising the implementation of the new regulatory framework across the European Union.
As negotiations continue throughout 2023, the AI Act is increasingly emerging as one of the most ambitious legislative initiatives ever undertaken in the field of artificial intelligence, with the potential not only to shape Europe’s digital future but also to influence how AI is governed around the world in the years ahead.