AI Legislation, By The Numbers
Key highlights this week:
We’re currently tracking 616 bills in 45 states (plus another 108 congressional bills) related to AI this year, 40 of which have been enacted into law.
Congress makes new recommendations for AI policy while a couple of deepfake bills earn committee approval at the federal level.
California’s governor wants to use AI to make his budget math work.
And sexual deepfake bills are on the move in Louisiana and North Carolina, potentially adding to the 18 states that have already put sexual deepfake laws in place.
A few weeks ago we added a counter to the top of these updates tracking the number of AI-related bills states have introduced and enacted so far this session. Last week that number topped 600 bills, and that excludes bills related to AI technologies like autonomous vehicles (another 77 bills). In the past few years, AI went from an idiosyncratic legislative interest to 150 mostly study bills in 2023 to over 600 and counting this year. This growing pile of studies, committee transcripts, and legislative language represent not only the interest of policymakers in regulating AI but also the speed at which the technology (its real and potential costs and benefits) has accelerated today. And I’m certain we’ve only scratched the surface.
Now that 33 states have adjourned their legislative sessions for the year, we want to take a step back and put some of these numbers into context. For instance, of the 616 bills introduced by state lawmakers, only 40 have been enacted thus far (a 6.5% enactment rate, which isn’t too bad for state legislation). A handful of others are sitting on governor’s desks ready to be signed into law, so that number will increase, but not by a significant amount.
Of those 40 enacted laws, only one could be classified as a “comprehensive AI law,” — Utah’s AI Policy Act, which we’ve dug into several times and is not nearly as broad as some of the more comprehensive bills we’ve seen debated this year but also has the distinction of getting signed into law and having recently gone into effect. In that category, Connecticut’s comprehensive bill died last week, while Colorado’s bill is awaiting the decision by Governor Polis (D) on whether it will become law this year. California is still debating several high-profile comprehensive AI bills, a legislative conversation that could go deep into the summer.
Seven of the 40 enacted bills (18%) were solely study bills, setting up new task forces to investigate the challenges AI will put before policymakers and make recommendations on initial moves. A handful of states got a head start on this type of work in 2023, but others are catching up this year. There were a handful of executive orders this year doing something similar from the governor’s desk. Another two bills were enacted solely for economic development purposes (such as this week’s Iowa law appropriating $2.8 million to expand certificate programs for AI, among other tech-focused subjects). Other enacted bills are even more vague, such as a Maryland law stating “the intent” to “evaluate the feasibility of creating a 3–1–1 portal utilizing artificial intelligence.”
The vast majority of AI bills, as we’ve detailed extensively, are bills addressing “deepfakes.” Twenty-five of the 40 enacted (63%) are deepfakes laws. Outside of study bills, deepfake laws are the most mature of the AI categories, with the first enactments taking place in 2019, picking up steam in late 2023, and representing a full onslaught of legislation in 2024 (297 bills introduced). Today, nearly half the states (23) have enacted laws to directly address sexual or political deepfakes or both.
The top states for AI-related bill introductions are California and New York. The Empire State enacted a deepfake law through its budget this year, and while Golden State lawmakers have yet to enact an AI-related bill this session, they have a long history of doing so (enacted 6 bills since 2018) and are currently debating some of the broadest comprehensive AI bills we’ve seen this year. The most effective state at enacting AI policy this year has been Utah (7 bills enacted this year), which we’ve held out as a potential middle ground for states looking to regulate AI.
But the jurisdiction with the most AI bills introduced this session is Congress (with 108 bills). None of which have been enacted into law. Despite plenty of press releases, committee gatherings, and reports issued, a divided Congress has yet to produce much substance on this emerging technology. In the meantime, state lawmakers are accustomed to moving (relatively) fast, and are always happy to fill any policymaking gaps left open by their colleagues at the federal level.
Recent Developments
In the News
OpenAI unveils GPT-4o: On Monday, OpenAI hosted an event introducing GPT-4o, a model with improved speed and interactive capability compared to the company’s flagship GPT-4. The model is “natively multimodal,” able to generate content or understand commands in voice, text, or images, and has a more natural-sounding voice that is able to mimic a wider range of emotions. The model is free to users, with paid subscribers able to access higher capacities.
Google I/O: On Tuesday, Google announced a laundry list of new AI tools and features at its annual developer conference Google I/O. The company featured demos with AI integrated into search, an AI agent that communicates with voice, and a text-to-video model.
Major Policy Action
Congress: On Wednesday, a bipartisan group led by Majority Leader Sen. Schumer (D) unveiled a roadmap for artificial intelligence policy that recommends spending $32 billion annually for research and development, and calls on narrow legislation, but avoids a comprehensive regulation of the industry. The proposal calls for a federal data privacy law and consideration of legislation to protect children from deepfake exploitation, a ban on the use of AI for social scoring, and content provenance information in AI-generated content. Additionally, the U.S. Senate Rules Committee advanced three bills on Wednesday to restrict the use of political deepfakes. The Protect Elections from Deceptive AI Act (S. 2770 ) would prohibit the distribution of materially deceptive AI to influence an election, while the AI Transparency in Elections Act (S. 3875) would require a disclaimer on political ads that have been substantially created or altered by AI. The Preparing Election Administrators for AI Act (S. 3897) would direct the Election Assistance Commission to develop guidelines for the administration of elections that address the use and risks of AI technologies.
California: Last Friday, Gov. Newsom (D) unveiled his proposal to cut the budget deficit that includes an 8% cut to government operations, in part through the use of AI. The state will partner with five companies to create generative AI tools to provide better services to the public, according to administration officials.
Colorado: Tech groups have begun to lobby Gov. Polis (D) on AI legislation (CO SB 205) that passed the legislature last week. Polis has a background in tech and was non-committal about the bill as it went through the legislative process. He has until June 7, 2024, to sign or veto the bill, or else it becomes law.
Louisiana: On Tuesday, the House amended and unanimously passed a sexual deepfake bill (LA SB 6), which would create the crime of unlawful dissemination or sale of images of another created by AI. The House amendments would expand exemptions to cover telecommunication providers and would need to be agreed upon by the Senate, which already passed the measure.
North Carolina: On Wednesday, the Senate passed a sexual deepfake bill (NC HB 591), which would modernize sex crime laws to include a realistic visual depiction created, adapted, or modified by technological means in provisions prohibiting sexual extortion. The Senate passed the bill, and it heads back to the House for approval of the changes made this week.
Notable Proposals
Illinois: On Tuesday, Rep. Janet Yang Rohr (D) introduced a bill (IL HB 5848), which would require that school policies on cyber-bullying also include the distribution of deepfakes of individuals engaged in an activity the individual did not engage in, including, but not limited to sexually explicit depictions. Police are investigating numerous incidents in an Illinois high school regarding images of students manipulated by AI.
New York: On Tuesday, Sen. Kristen Gonzalez (D) introduced a bill (NY SB 9381), which would subject proprietors of chatbots that provide misleading information to liability. Liability would stem from any materially misleading, incorrect, contradictory, or harmful information to a user that results in financial loss or other demonstrable harm to the user and could not be disclaimed simply by disclosing the fact the user is interacting with a chatbot. She also co-sponsored a bill (NY SB 9401) that would require businesses to conduct impact assessments for AI systems with a surcharge on employers whose AI systems reduce the number of worker hours or lead to layoffs.
New Jersey: On Thursday, lawmakers filed two bills to study AI and its impacts on the state. One bill (NJ AB 4399) would require a study of the impact of state agencies procuring, implementing, and operating AI. Another (NJ AB 4400) would establish the New Jersey Artificial Intelligence Advisory Council.