Legal Status of the AI-Transparency Duties at the Consumer-Facing Interface
Through a structured comparative analysis of eleven jurisdictions, this paper categorizes diverse AI-transparency duties at the consumer-facing interface into a four-level typology, revealing partial convergence in regulatory techniques while highlighting significant variations in their legal force and implementation across the European Union, China, the Republic of Korea, and other global regions.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the digital world, the screen you look at is more than just a window; it is the point where a human hand meets a machine's mind. For decades, the rules governing what happens on that screen have been scattered across different legal categories, like privacy notices or terms of service. But the rise of artificial intelligence has changed the nature of that meeting. When a computer program writes a story, diagnoses an illness, or acts as a friendly companion, the line between human and machine blurs. The question for lawyers and policymakers is no longer just about the technology itself, but about what the user sees and knows at the moment of interaction. If a person is talking to a bot that sounds exactly like a human, or reading a news article written by a machine, they have a right to know. This is the realm of transparency: the duty to make the invisible nature of a machine visible to the person using it.
A team of researchers set out to map how eleven different countries and regions are handling this duty. They looked at the European Union, the United States, China, the United Kingdom, Canada, Japan, South Korea, Singapore, Australia, Brazil, and the United Arab Emirates. Their goal was not to count how many laws exist, but to understand what specific actions are required on the screen itself. They focused on four main types of visible duties: telling a user they are talking to a machine, labeling content created by a machine, warning users when a machine makes a high-stakes decision, and making sure the organization behind the machine is open about its rules. By treating the screen as a common point of comparison, they could see how different legal systems, with their own unique histories and languages, are trying to solve the same problem.
The researchers found that while the laws are written in very different ways, the actual requirements on the screen are starting to look similar. In the European Union, China, and South Korea, the rules are strict and written into binding law. These places require companies to clearly label AI-generated images or text and to tell users immediately when they are interacting with a system rather than a person. For example, in China, specific regulations mandate that AI-generated content must carry a label, and in South Korea, companies must publish information about how they manage risks for their high-impact AI systems. The European Union has established a comprehensive set of rules that apply across its member states, requiring disclosures for everything from chatbots to deepfake videos.
In contrast, the United States does not have a single national law covering all of this. Instead, the rules are a patchwork of federal enforcement actions and laws passed by individual states. The federal government uses existing consumer protection laws to stop companies from lying about their AI, while states like California, New York, and Colorado have passed their own specific laws. These state laws often focus on "companion" chatbots—systems designed to be friends or confidants—requiring them to disclose their non-human nature and have safety protocols for users in crisis. This creates a situation where a company operating across the US must navigate a complex map of different state requirements, rather than following one national standard.
Other regions are still in the process of forming their rules. Countries like Canada and Brazil have proposed laws that look similar to the European model, but these have not yet become active law. Singapore and Australia are currently relying on voluntary guidelines and industry standards, encouraging companies to be transparent without forcing them with strict penalties. The United Kingdom and the United Arab Emirates are taking a different approach, using existing data protection laws or new government bodies to oversee AI, rather than creating a single, dedicated AI law. The researchers noted that in these places, the duty to be transparent often depends on the specific context, such as whether the AI is making a decision about a person's job or credit.
The study concludes that while the legal foundations are fragmented, the practical result on the screen is converging. Across the world, the recurring demand is for the same four things: a clear notice that a machine is involved, a label on synthetic content, a warning for high-risk decisions, and public information about how the system is governed. However, the researchers warn that simply putting a label on a screen is not enough. A generic warning that says "this is AI" might be visible, but it does not explain who is responsible, what risks are involved, or how a person can challenge a decision. The real challenge for the future is to ensure that these visible disclosures connect to the actual functions of the machine and the people who control it. As AI becomes a standard part of daily life, the goal is to move beyond simple labels to a system where the user can truly understand the relationship between their actions and the machine's responses.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.