Can human rights law keep up with artificial intelligence?

The importance of a broad view on human rights protection 

According to Ezra de Leeuw, PhD candidate at Erasmus School of Law, the timing is no coincidence. "Our article is particularly relevant now that AI is increasingly being adopted in both public and private sectors, and the societal impacts of AI are becoming more apparent," De Leeuw says. The researcher points out that legislation like the AI Act is a meaningful step forward, but adds a caveat: it "might still have blind spots that leave certain aspects of human rights protection incomplete." One reason for those blind spots, De Leeuw explains, is that much of the existing research looks at a single human right in isolation. The study aims to offer the broader view that is still missing, one that helps researchers and policymakers understand AI's impact on human rights protection as a whole. 

The debate leans heavily toward risk 

One of the study's clearest findings is a striking imbalance. Of the 128 articles reviewed, 112 studies identify negative effects of AI on human rights, while only 49 studies point to positive effects. For marginalised communities, the gap is even starker. Roughly a third of the articles describe AI harming these groups, while none report AI benefiting them. 

"This particular finding suggests that the current academic debate is primarily focused on the risks and challenges for human rights," De Leeuw says. The researcher sees this as a sign of growing awareness rather than pure pessimism, calling it "a positive step toward greater attention." At the same time, De Leeuw notes a simple explanation for the pattern: "legal scholars, like lawyers in general, tend to examine harms more closely than benefits and opportunities." 

Old problems made worse, and entirely new ones 

The researchers distinguish between two types of harm. AI can reinforce existing problems, making them worse or harder to detect. It can also create new issues that simply did not exist before. 

De Leeuw offers a concrete example of the first category: automated content moderation. "If AI systems are used to detect harmful content on social media, the AI systems might wrongfully remove posts from certain communities or about certain topics that are actually not considered to be harmful." That can limit the visibility of certain groups and restrict freedom of expression. 

Also, with regard to new issues that arise, De Leeuw points to deepfakes. "This novel application of AI can infringe people's privacy, as the AI application uses personal data of the person whose pictures are used," De Leeuw says. It is a form of harm that simply did not exist before AI made it possible. 

Who is actually responsible? 

Human rights law has traditionally held states responsible for protecting citizens, including from harm caused by private companies. But De Leeuw argues this model struggles with AI for several reasons. "The sector is dominated by a small number of powerful corporations whose influence can undermine regulation, even at the EU level. AI systems often operate across borders, so a single social media algorithm can affect people in many countries within different legal systems. And many AI models function as "black boxes," making it nearly impossible to understand how a decision was reached, let alone challenge it in court.” 

If tech companies could be held accountable more directly, De Leeuw says, "it could provide citizens with stronger protection against AI-related harm," giving companies a real incentive to respect human rights obligations. 

When there is no single victim 

Another core problem is that human rights law is centred around individual harm. To bring a case before the European Court of Human Rights, a person typically has to prove they personally suffered concrete harm. But AI-driven harms, especially mass surveillance or biased algorithms, often affect entire groups at once, without a clearly identifiable individual victim. 

"This collective nature of AI-driven harm complicates the process of filing individual complaints and proving direct harm," De Leeuw explains. One possible solution already exists under Dutch law: collective legal action, which allows an organisation to seek redress on behalf of a larger group. However, De Leeuw notes that this route remains significantly limited within the case law of the European Court of Human Rights. The underlying issue is that the Court still generally requires claimants to prove personal, concrete harm. It has occasionally made exceptions in specific mass surveillance cases, but researchers describe this as a narrow workaround rather than a genuine shift toward recognising groups as rights holders in their own right. 

What can worried citizens do? 

For people concerned about their rights in an AI-driven world, De Leeuw's advice is to stay informed. Public and academic attention to AI's risks has already helped drive new regulation offering some degree of protection. NGOs play an important role in monitoring compliance and supporting affected individuals, and the Council of Europe's Framework Convention on Artificial Intelligence may offer further safeguards in the years ahead.

A collaborative effort 

The publication itself reflects a broader collaboration within Erasmus School of Law and the Erasmus Centre of Law and Digitalization, with support of the Sectorplan SSH-Breed on Digitalisation, bringing together five researchers actively working to shape this emerging field. As Quintavalla puts it, the project "laid the groundwork for continued research in this critical field," with more work at the centre already underway.