The Three Phases of State AI Regulation
In the busiest week for AI legislation so far, state lawmakers passed over a dozen AI bills out of their legislative chamber of origin as key crossover deadlines approached. Some key highlights:
Utah legislators sent the AI Policy Act along with an electoral deepfake bill to Gov. Cox for his signature to become law.
Deepfake bills in Indiana and Washington also passed their legislature and were sent to the governor.
Lawmakers in Alabama, Arizona, Florida, Idaho, Kentucky, Oklahoma, Oregon, West Virginia, and Wyoming all passed deepfake bills out of their chambers of origin, sending them to the second legislative chamber for their consideration.
We’re approaching the halfway point in many of the 42 state legislatures currently in session. One indication of that is the crossover deadlines approaching in many of these states. That means bills need to be passed out of their legislative chamber of origin or they’re dead for the year. That’s one reason we saw over a dozen AI bills passed out of a chamber this past week. But it’s also because the AI issue has just exploded onto the scene for state legislators. As of this morning, we’re tracking over 500 AI-related bills in the states and it’s a good time to take a step back and evaluate the landscape.
The progression of AI bills has come in three stages: study bills, narrowly focused bills, and comprehensive legislation. Right now, we’re moving out of the study phase, are deep in the narrowly focused phase, and are just beginning the comprehensive phase. Let’s dig into each of these.
First, we had the study bills. In 2023, a majority of the 150 bills introduced on AI set up task forces and committees to study AI and report back to the full legislature on recommendations for regulating the technology. Those efforts are still ongoing and more study bills are being introduced and moved each week. Governors are also taking direct action by setting up study committees via executive orders and some state legislatures have assigned the task of examining AI in detail to already established standing committees. Currently, 27 states are actively studying AI policy and we’re tracking 70 additional bills to establish study organizations. But while states began to study the topic in 2023, this year they have moved on to taking direct action.
Second, lawmakers are now addressing specific problems through narrowly focused bills. This year, that focus is concentrated on deepfakes. The two headline-grabbing issues that states have begun to address are political deepfakes and sexual deepfakes. Already, there has been a much-publicized incident of deepfake use in a political campaign with an AI-generated robocall in the Democratic presidential primary in New Hampshire, and with elections fast approaching, lawmakers are very concerned with deepfake media imitating candidates to deceive voters. Sexual deepfakes are another issue of grave concern after several incidents at high schools and the recent proliferation of deepfake images of popstar Taylor Swift spread online. In total, we’re currently tracking 227 state bills related to deepfakes. Other bills related to generative AI include legislation combating fraud through impersonating voices or video (NJ AB 3912), protecting someone’s likeness such as a performer’s look or voice (MS SB 2778), and requiring watermarks to clearly identify “synthetic media” (WY SF 51). As new controversies surrounding AI use emerge, this list of narrowly focused bills will expand. We’re solidly within this phase with more than a dozen deepfake bills passing a legislative chamber just this week.
Finally, we’re now entering the third phase of AI legislation in the states. As we’ve highlighted over the past few weeks (read out analyses of CA SB 1047 & CT SB 2), lawmakers are taking what they’ve learned in their study committees and putting those lessons together into comprehensive legislation to set a framework on how AI is developed, deployed, and maintained in the states. One difference between these comprehensive bills and the more narrowly focused bills is that the comprehensive bills are proactively anticipating issues instead of reacting to them as they occur. Another differentiator is that the comprehensive bills are not only addressing the end use of an AI tool but also the system and the model that runs the AI tool. These bills address broader issues such as the safety of AI models, bias, data privacy, notice and transparency, and opt-outs. We’re still at the beginning of this phase (we’re tracking 15 bills today that we consider comprehensive) but it could prove to be the most vital aspect of state involvement in regulating AI. For now, Congress does not look ready to act on AI, but if states begin to enact comprehensive regulatory frameworks aimed at the development and deployment of major AI models, it could spark a federal response.
We’re still in the early innings of AI development and regulation. It seems like each week brings us major news. We here are multistate.ai will be here to connect all the fast-moving dots and make sense of the actions states are taking to regulate AI and how those actions fit into the AI story at large.
Recent Developments
In the News
Microsoft partners with Mistral and Gemini’s Faceplant: On Monday, Microsoft announced a new partnership with French AI startup Mistral AI, expanding the tech giant’s AI partnerships beyond OpenAI, which immediately got the attention of European regulators. The fallout from controversial image generation produced by Google’s Gemini AI model continues this week as critics test the “wokeness” of the model’s text generation, with Alphabet CEO calling the situation “completely unacceptable.”
Major Policy Action
Utah: The House passed two bills related to AI, both of which the Senate had previously approved, sending those bills to the governor for his signature to become law. Last Thursday, the House passed an electoral deepfake bill (UT SB 131) requiring a disclosure for political advertising containing synthetic media. And on Wednesday, the House passed a broader “AI Policy Act” bill (UT SB 149) that sets up various offices dedicated to AI and would establish liability for the use of AI that violates consumer protection laws if not properly disclosed.
Indiana: On Tuesday, the Senate passed two deepfake bills, one targeting political deepfakes and the other sexual deepfakes, that the House had previously passed, sending both measures to the governor for his signature to become law. The sexual deepfake bill (IN HB 1047) would define certain images created by AI to constitute an "intimate image" for purposes of the crime of distributing an intimate image. And the political deepfake bill (IN HB 1133) would prohibit the use of deepfakes in political advertising without a disclaimer and provide a civil action for the person being depicted.
Washington: On Wednesday, the Senate passed a sexual deepfake bill (WA HB 1999), which the House had already approved, sending the measure to the governor for his signature to become law. The bill adds criminal provisions for the fabrication, distribution, possession, or viewing of sexual deepfakes of minors and adds criminal provisions for disclosing a sexual deepfake of a person without consent.
Oklahoma: Last Wednesday, the House passed a deepfake bill (OK HB 3073), sending it to the Senate for consideration. The bill would criminalize the dissemination or publication of a digitization depicting another person's name, image, or likeness without their written consent and with the intent to cause emotional, financial, or physical harm.
Alabama: Last Thursday, the House passed a deepfake bill (AL HB 161), sending it to the Senate for consideration. The bill would make it unlawful to knowingly create, record, or alter a private image without consent, including artificially-generated images if a reasonable person would believe it actually depicts an identifiable individual.
Oregon: On Monday, the Senate passed a political deepfake bill (OR SB 1571), sending it to the House for consideration. The bill would require a disclosure of the use of synthetic media in campaign communications.
Idaho: On Tuesday, the House passed a deepfake bill (ID HB 575), sending it to the Senate for consideration. The bill would make it unlawful to disclose explicit synthetic media without consent if the disclosure would cause the identifiable person substantial emotional distress, or to harass, intimidate, or humiliate a person or obtain money through fraud.
Wyoming: On Tuesday, the Senate passed a deepfake bill (WY SF 51), sending it to the House for consideration. The bill would require the dissemination of misleading synthetic media to display a clear and conspicuous notice identifying the synthetic media.
West Virginia: After the Senate passed two sexual deepfake bills last week, the House passed two of their own deepfake bills on Wednesday, sending them to the House for consideration. A sexual deepfake bill (WV HB 5516) would prohibit the use of deepfake images for the criminal invasion of privacy or the unlawful depiction of nude or partially nude minors or minors engaged in sexually explicit conduct. And a political deepfake bill (WV HB 4963) would prohibit the use of AI to attribute remarks to a political candidate that were never made within 90 days of an election.
Florida: On Wednesday, the House passed a political deepfake bill (FL HB 919) sending it to the Senate and the Senate passed its own broader deepfake bill (FL SB 1680), sending it to the House. The broader Senate bill would require watermarks for deepfake media, disclosure for AI-generated political ads, and prohibits the use of AI in obscene material depicting a minor. The narrower House bill would require a disclaimer for political advertisements created in whole or in part with the use of generative artificial intelligence.
Kentucky: On Wednesday, the House passed a deepfake bill (KY HB 45), sending it to the Senate for consideration. The bill would make it unlawful for a person to willfully and knowingly disseminate a deepfake of a depicted individual without consent. And last Friday, the Senate passed a political deepfake bill (KY SB 131), sending it to the House for consideration.
Arizona: On Thursday, the Senate passed a political deepfake bill, sending it to the House a week after the House sent its own bill on the same topic to the Senate. The House version (AZ HB 2394) would provide for a cause of action by a candidate for public office for a digital impersonation without consent and without disclaimers. While the Senate version (AZ SB 1359) would prohibit the distribution of a deceptive and fraudulent deepfake of a candidate or political party within 90 days of an election without a disclaimer.
Notable Proposals
Louisiana: On Wednesday, lawmakers introduced a bill (LA SB 118) that would require the developer of an AI foundation model to register the model with the state. The bill defines a foundation model as “a large-scale machine learning model trained on a broad data set that can be adapted for multiple tasks.”
New York: On Wednesday, Assemblyman George Alvarez (D) introduced a pair of bills aimed at regulating automated decision making tools in employment. One bill (NY AB 9314) would require a disparate impact analysis to be conducted annually on such tools, while another bill (NY AB 9315) would restrict the use of such tools and employee monitoring tools, and require bias audits that are shared with employees and the state.
Rhode Island: On Thursday, lawmakers introduced a bill (RI HB 7786) that would require impact assessments and risk management programs for consequential AI decision systems. The measure would require developers and deployers to self-certify with enforcement through the attorney general’s office.