Bill to add AI disclaimers appears ready to pass this session

After Meta’s AI chatbot fabricated sexual harassment allegations against state officials, lawmakers are poised to pass legislation that would require companies to start including warnings that their AI outputs may be wrong.

City & State reported in April that Meta AI consistently made up false stories of sexual harassment allegations against sitting lawmakers. The information was completely false, with the chatbot offering varying levels of details about the allegations, including at least one fake name for a supposed accuser.

In response to the reporting, state Sen. Kristen Gonzalez and Assembly Member Clyde Vanel introduced legislation to ensure that the chatbots provide notice that results “may be inaccurate and/or inappropriate.” Companies that don’t comply would face stiff penalties of up to $100,000. According to Bloomberg Government, New York would be the first state in the country to enact such a regulation.

Vanel introduced the legislation on May 3, with Gonzalez following on May 15. Despite the late timing, the sponsors expect the bills to pass before the session ends next week. Vanel said the legislation garnered swift support, noting the fact that Meta AI was specifically making up harassment allegations about lawmakers. “I’m just relieved that it has so much support and that we have the opportunity to pass this bill this session,” he said.

After being alerted to the problem, Vanel’s office conducted its own review of eight commonly used AI chatbots, finding that only half offered consistent warnings that the outputs may be wrong. He said his legislation is meant to address the issue without hindering the progression and learning models of developing technologies. “I want developers to develop new types of systems, I want them to get smarter and better,” Vanel said. “But I also want people to understand that they’re not perfect.” The legislative language does not include specifics about what those warnings should look like, how long they remain on the screen or how often the chatbot should issue them. Vanel said those are questions that he expects will get hashed out in the regulations. But he added that based on conversations he has with chatbot owners, he does not expect issues with compliance.

The state Senate was originally scheduled to vote on its version of the legislation on Thursday, but Gonzalez told City & State that it was laid aside due to a last-minute amendment made to the Assembly version. Both chambers need to pass identical versions of a bill in order for it to become law, and the May 28 tweak in the Assembly took them out of alignment. But both Gonzalez and Vanel are confident that will be remedied soon so that the bills can move forward.

The new legislation is not the only swift government response to Meta AI’s fabrications. Soon after the results were reported, state Attorney General Letitia James sent a letter to Facebook’s parent company looking for answers about the false allegations its chatbot was making.