As AI systems become more complex, scholars are racing to develop legal frameworks. Here are just a few examples.
Depending on the context, AI tools could be viewed as products, services, autonomous agents or entities that may someday warrant some form of legal personhood. The debate is playing out across the globe.
Can an artificial intelligence tool, such as a chatbot, be held responsible if people committed crimes based on information they got from the tool?
Should chatbots, which are programmed to closely mimic human thought patterns and are increasingly treated as companions, be subject to legal checks that acknowledge those tools’ developing independence or autonomy?
These are not easy questions, but ones lawyers are grappling with across the globe, from a handful of cases in the U.S. to regulations being considered in Europe.
As AI agents grow more powerful, governments and regulatory bodies are also weighing these questions. In response, some are devising new legal frameworks to govern the emerging technology in ways that anticipate a continued evolution in capabilities — perhaps beyond what humanity can imagine. Where they land will ultimately influence how AI is regulated, experts say, and could determine who is to be held responsible when systems discriminate, cause harm or lead people to break the law.
While the question of criminality is being argued, at the heart of the debate is something much more basic.
“The fundamental question is: What is a chatbot?” said Patricia Williams, university distinguished professor of law and humanities, who has been writing about the ethical and philosophical scopes of AI in recent years.
Depending on the context, AI tools could be viewed as products, services, autonomous agents with some ability to reason and plan or — in more speculative corners of the debate — entities that may someday warrant some form of legal personhood. Each of those classifications carries different implications for legal liability and accountability, experts say.
“When they’re categorized simply as products — something provided by a company — it’s a very different relationship than if we put it in the realm of services, or human services,” Williams said. “Services can be governed … by things like civil rights laws, fairness laws and questions of accountability in a very different way than simply a product that, say, got broken.”
Today, so-called agentic AI systems, which, unlike chatbots, have some capacity to act independently, are widely viewed in legal circles as “inanimate” objects, lacking consciousness, subjectivity or self-awareness, experts say.
But the challenge, according to Williams, lies with the fact that governance and regulation are inherently at odds with the proprietary posture of many AI companies and systems. Companies often resist outside oversight and develop their systems in a so-called “Black Box,” meaning their inner workings are largely secret.
Editor’s Picks
In the U.S., the debate is starting to touch that more future-looking notion of “AI personhood,” which would entail attributing the rights, protections or legal standing currently reserved for “persons” to chatbots and other tools, Williams said. That’s because, in legal philosophy, a “person” isn’t necessarily a human being, but any entity capable of holding rights or responsibilities recognized by law, experts say.
In Ohio, for example, lawmakers are attempting to preempt any broader efforts at assigning personhood to AI by preemptively declaring such systems as “nonsentient.” States like Idaho and Utah have also introduced similar measures opposing the classification of AI systems as legal persons.
In California and Colorado, lawmakers have focused less on personhood and more on regulating how AI systems are deployed, particularly as it pertains to employment, housing, health care and discrimination.
In Europe, regulators have generally taken a more expansive approach to AI governance, focusing on privacy protections, transparency around the algorithms that companies use and accountability for AI platforms if something goes wrong.
But there have also been some creative and niche offshoots in the classification debate: some legal scholars, for example, are reaching all the way back to Roman law and how slaves’ behavior rarely held consequences for their masters, as a means of more deeply understanding the “accountability gap” between AI systems and their human creators and operators.
Europe has emerged as the world’s most aggressive regulator of digital privacy largely because privacy protections are explicitly outlined in the constitutional framework of the European Union, said Hilary Robinson, an associate professor of law and sociology at Northeastern University’s School of Law.
The EU’s sweeping data privacy and security law known as General Data Protection Regulation, or GDPR, is the clearest example of Europe’s regulatory approach to AI technologies. Since it was enacted in 2018, the law has allowed European regulators to impose strict rules on global technology companies and establish mechanisms such as the “right to be forgotten,” which, under certain circumstances, allows individuals to request the removal of personal data from online platforms, Robinson said.
With GDPR, it’s harder to have an “accountability gap.” Under that law, Robinson said, large online platforms and data intermediaries are treated not merely as hosts for information but as entities with affirmative legal responsibilities for how user data is collected, stored and processed.
American privacy law is fragmented and comparatively weak, she said. The U.S. lacks any comparable constitutionally-enshrined right to privacy. Additionally, the nation’s privacy protections have instead been “cobbled together” through court decisions interpreting the Fourth Amendment and the due process clauses of the Fifth and 14th Amendments, Robinson said.
“It’s kind of judge-made doctrine,” Robinson said. “But the EU wrote the right to privacy directly into its constitutional structure, which allowed it to implement those protections statutorily through the GDPR.”
As AI systems continue to endlessly generate content — everything from songs and novels to patentable inventions — longstanding assumptions about authorship, ownership and legal accountability are beginning to fracture, experts say.
Ursula Smartt, an associate professor of law and criminology based in London, said that courts and lawmakers are increasingly confronting a foundational legal problem: under most existing U.S., U.K. and European legal frameworks, only human beings can qualify as inventors or authors entitled to patent and copyright protections.
“AI-generated copyrighted works, such as plays, songs and books, cannot have their own copyright,” Smartt said, adding that global patent laws stipulate that “the inventor must be a human.”
“In other words, they must be sentient,” she said.
