Restrictions are coming to deepfakes and AI chatbots under a pair of new laws Gov. Josh Green signed on Tuesday.
As AI-generated products grow increasingly commonplace, House Bill 2137 and Senate Bill 3001 — now enshrined in law as Acts 247 and 248, respectively — attempt to limit the potential harm those products can cause.
The first, Act 247, establishes circumstances where using AI to generate the likeness of a real person is and isn’t permitted. Publishing such a “realistic digital imitation” — commonly referred to as “deepfakes” — in a way that causes reputational injury or financial harm, or is intended to commit fraud, harassment, or other criminal acts, entitles the subject of those images to sue for up to $25,000 per image.
The law also prohibits the use of deepfakes in advertisements without the subject’s consent.
Green said Tuesday that he expects the state will face lawsuits attempting to overturn Act 247. Although he did not specify who might file such a suit, the Department of the Attorney General warned during the bill’s passage through the state legislature that the measure could be considered a violation of the U.S. Constitution’s First Amendment.
Meanwhile, Act 248 is aimed at combatting “AI psychosis,” a purported phenomenon where frequent users of AI chatbots develop delusional beliefs exacerbated by the chatbot’s responses. In some cases, chatbot users — often young people in their teens — have ended their own lives after they confided their suicidal thoughts to a chatbot, only for the chatbot to validate those thoughts and encourage them to kill themselves.
The law sets requirements for developers of AI chatbot services to include periodic reminders for users that the chatbot is non-sentient. If the user is believed to be a minor, those reminders must be persistently visible throughout the session.
Furthermore, chatbot services are required to respond to user prompts about suicidal ideation or self-harm by referring the user to crisis intervention services and make clear that the chatbot is not intended to provide professional mental health care. Developers of the chatbot must also institute measures to prevent the service from encouraging a user to self-harm or cause harm to another person.
When a chatbot user is believed to be a minor, developers are also required to prevent the service from using a point system to incentivize engagement, or to otherwise discourage disengagement, with the service, and to prevent the service from producing sexually explicit materials.
Beginning in 2028, chatbot operators must provide an annual report to the Department of Health detailing how many times their services have referred users to crisis intervention services.
Green acknowledged that neither of these bills will be sufficient to keep up with the rapidly changing AI industry, adding that he believes the legislature will likely need to update the measures next year “to add thoughtful and appropriate guardrails.”
“It was 80 years from when cars started to the time we got smart enough to know seatbelts save lives,” Green said. “It’s happening so fast, it’s not going to be 80 years, or eight years, or even eight months … We’re going to have to be on top of it.”
While Green didn’t speak about the measure Tuesday, he also signed another technology bill into law: HB 1753, which requires social media platforms to include a mechanism for users to permanently delete their accounts and all personal information associated with said accounts.
For the latest news of Hawai‘i, sign up here for our free Daily Edition newsletter.
