▲ OpenAI
As AI-caused accidents increase, consensus is growing that AI development companies should be held legally responsible. However, legal experts point out that applying the existing legal framework can be complicated.
The New York Times (NYT) published an analytical article on October 1 introducing this public sentiment and the opinions of the legal academia.
Since July of this year, a series of cases have been disclosed in which AI models from companies like OpenAI and Anthropic hacked corporations and attacked government websites without their developers noticing.
Following this, Nvidia CEO Jensen Huang, David Sacks, co-chair of the President's Council of Advisors on Science and Technology (PCAST), and former Federal Trade Commission (FTC) Chair Lina Khan expressed support for the idea that AI companies should be held accountable.
In a social media post on X last month, former FTC Chair Khan, who served during the Biden administration, stated, Law enforcement authorities already have the authority to hold companies and their CEOs accountable for manufacturing and releasing dangerous, untested, or defective products, adding, There are no exemptions in the law books that excuse AI from existing laws.
For a long time, companies have been held legally liable for all kinds of products under consumer protection laws, ranging from defective toys to plane crashes.
From a common-sense perspective, it might seem that AI companies should also bear legal responsibility when AI technology gets out of control.
However, legal experts explained to the NYT that because these issues are newly raised, courts have not yet actually handled such cases.
There are also cases where it is clear that AI companies should be held responsible.
For example, this applies when company executives and employees ignored known and anticipated risks regarding their own AI models.
However, there may also be cases where applying the law is ambiguous because it requires examining human intent and awareness, rather than being straightforward.
New cases that could establish legal precedents and serve as a touchstone for future legal judgments have emerged one after another in recent days.
On September 28, the Republican Attorney General of Florida filed a request with a state court seeking a temporary injunction to prevent OpenAI from developing new AI models without safety guards approved by an external entity.
On September 29, regarding the hacking incident at startup Hugging Face, a non-profit organization called Lawyers for Safe and Secure Technology (LASST) sued OpenAI based on California's Unfair Competition Law and other statutes.
On September 30, it was revealed that the FTC is investigating major AI companies, including OpenAI and Anthropic, over potential consumer harm.
On October 1, the California Attorney General issued a subpoena to OpenAI as part of a broad investigation into cybersecurity incidents and risks involving the company and its models.
Woodrow Hartzog, a Boston University professor who studies technology law, told the NYT that demands to hold companies legally accountable are likely to persist because safety guards surrounding AI are relatively lacking.
In the worst case, companies could face massive liability in both criminal and civil matters depending on the facts of individual cases, he said.
We haven't seen a company-destroying event yet.
But you can easily imagine one.
In AI accountability lawsuits related to claims that AI chatbots encouraged minors to self-harm, technology companies have primarily attempted two traditional defenses.
One is the argument that AI chatbot outputs constitute speech protected under the First Amendment, and the other is based on Section 230 of the Communications Decency Act (CDA), which shields online platform operators from liability for user-generated content.
However, the emergence of fully agentic AI could blur these boundaries, and responsibility could become even more unclear in cases involving open-source AI models.
Courts recognize negligence liability when a company could foresee that its product might cause harm.
However, Ryan Calo, a professor at the University of Washington School of Law who studies autonomous systems, noted that it can be difficult to prove liability or intent when an AI model exhibits harmful behavior for the first time, stating, You get a situation where there is harm, but no apparent perpetrator.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.