(Seoul = Yonhap News) Reporter Lim Hwa-seop = Accidents caused by artificial intelligence (AI) are increasing. There is a growing consensus that AI development companies should be held legally responsible, but some legal experts point out that applying the existing legal system may be difficult.
The American daily New York Times (NYT) published an analysis article on the 1st (local time) introducing this trend of public opinion and the opinions of legal scholars.
Since July of this year, cases of AI models such as OpenAI and Antropic hacking companies and attacking government websites without developers knowing have been revealed one after another.
Afterwards, Nvidia CEO Jensen Huang, U.S. Presidential Council of Advisors on Science and Technology (PCAST) co-chairman David Sacks, and former Federal Trade Commission (FTC) Chairman Lina Khan expressed their support for the idea that AI companies should be held accountable.
Former FTC Chairman Khan, who served during the Biden administration, said in an
Consumer protection laws have long made companies liable for everything from defective toys to airplane crashes.
Common sense suggests that AI companies may also be held legally responsible if AI technology gets out of control.
However, legal experts reported by the NYT explain that because these issues are newly raised, the courts have not yet actually dealt with such cases.
There are also incidents where it is clear that AI companies must be held accountable.
For example, there was a case where AI company executives and employees ignored the 'expected risks' posed to their AI model even though they were aware of them.
However, there may be cases where it is unclear to apply the law, as it is not a clear case, but it is necessary to consider what the person's intention was and how much he or she knew.
New cases that have the potential to set precedents and become a touchstone for future legal decisions have been filed one after another in recent days.
On the 28th of last month, Florida's Republican attorney general filed a request with the state court to temporarily block OpenAI from developing new AI models without safety measures approved by an external agency.
On the 29th of last month, the non-profit organization 'Law Guardians for Safe Science and Technology' (LASST) filed a lawsuit against OpenAI based on California's unfair competition law in connection with the hacking incident of the startup 'Hugging Face'.
On the 30th of last month, it was revealed that the FTC was conducting an investigation into the possibility of consumer damage against major AI companies such as OpenAI and Antropic.
On the 1st of this month, the California Attorney General issued a subpoena against OpenAI as part of a "broad investigation into cybersecurity incidents and risks involving the company and its models."
Woodrow Harzog, a Boston University professor who studies technology law, told the NYT that calls for legal liability are likely to continue because of the relative lack of safeguards surrounding AI.
He said, "In the worst case, depending on the facts of each case, you may be subject to enormous liability in both criminal and civil terms. There has not yet been an incident that is serious enough to bring down a company. However, such an incident is fully imaginable."
In AI liability lawsuits involving claims that AI chatbots encouraged teenagers to harm themselves, technology companies have primarily attempted two traditional defenses.
The claim is based on the claim that the output of the AI โโchatbot constitutes โspeechโ protected under the First Amendment of the Constitution and Article 230 of the Communications Decency Act (CDA), which allows online platform operators to be exempt from liability for content posted by users.
However, with the advent of fully agentic AI, the lines may become blurred, and liability may become even more unclear, especially in cases involving open source AI models.
Courts recognize negligence liability when a company could have foreseen that its product could cause harm.
However, Ryan Calo, a professor at the University of Washington Law School who studies autonomous systems, said that it may be difficult to prove responsibility or intent to harm when an AI model exhibits harmful behavior for the first time. โIt becomes possible for a situation where there is no perpetrator but there is a victim,โ he said.
