It was a chilly Sunday for much of New Hampshire, and voters in The Granite State were preparing for the state’s storied primary elections for the 2024 cycle. In just two days, people would be heading to the polls to cast their ballots in what was shaping up to be a historic race.
The New Hampshire electorate takes much pride in its famed reputation as the home of the “first-in-the-nation” primary and the perennial kingmaker for presidential nominees. New Hampshirites are also no stranger to a little bit of political drama — in fact, some may say they embrace it. And this primary was no different.
On the Republican side, some predicted that a moderate electorate might give Nikki Haley an opening to gain momentum or even score a major upset victory against frontrunner Former President Donald Trump. Meanwhile, Democrats found themselves in a peculiar situation: incumbent President and almost-certain nominee Joe Biden was not listed as a candidate on New Hampshire ballots due to a disagreement over the state’s placement on the primary calendar. This meant that to prevent a potentially embarrassing loss, the President would have to successfully turnout voters for a write-in campaign.
That Sunday, however, something unexpected happened.
Voters began receiving calls with a voice that sounded remarkably like President Biden – but discouraging people to go to the polls. “Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again,” the AP reports the call as saying, “Your vote makes a difference in November, not this Tuesday.” The call was a deepfake, an AI-generated audio designed to mimic President Biden’s speech. And it may be a warning sign of how artificial intelligence could be used as a tool for spreading misinformation during this and future elections.
States Spearhead AI Legislation
Recent years have seen the rapid growth and increased sophistication of artificial intelligence technologies. With the release of OpenAI’s ChatGPT to the public sphere in 2022, the everyday use of AI (for better or worse) has become evident to the broader population. Increased public awareness of the implications of AI has led to increased attention from governments, particularly at the state level.
Collage by Jogos Public Assets via Wikimedia Commons
Over the course of just one year, there has been a marked increase in states with viable legislation focused on AI. According to the National Conference of State Legislatures, in 2023 “at least 25 states, Puerto Rico, and the District of Columbia introduced artificial intelligence bills.” For 2024, that number has increased to 40 states, plus “Puerto Rico, the Virgin Islands and Washington, D.C.” In the case of many states, increased interest in AI has meant increased resources directed towards studying AI-related issues. The Brennan Center for Justice reports that a number of states, “at least 12 — including Alabama, California, Colorado, Connecticut, Illinois, Louisiana, New Jersey, New York, North Dakota, Texas, Vermont, and Washington,” as of 2023, “enacted laws that delegate research obligations [regarding AI] to government or government-organized entities.” And that list of states continues to grow, with Indiana and West Virginia creating an AI-focused task force and select committee, respectively, in the 2024 legislative session.
Georgia has by no means been an exception to this. In the summer of 2023, a House Subcommittee on Artificial Intelligence was created, chaired by Representative Brad Thomas (R-Holly Springs). The 2024 legislative session has seen a number of AI bills introduced and receive attention.
AI Legislation Takes Shape in Georgia
Three AI bills introduced in the 2024 legislative session caught particular attention: H.B. 986, 988, and 1361. Each bill was sponsored by Thomas.
H.B. 988 requires an annual inventory of how AI is being used by state agencies and has the Georgia Technology Authority create an annual report of this inventory. The bill states that each inventory must collect the following information: “The name of the system and the vendor, if any, that provided such system… description of the general capabilities and uses of such system… whether such system was used to independently make, inform, or materially support a conclusion, decision, or judgment; and… whether such system underwent an impact assessment prior to implementation.” H.B. 1361, meanwhile, seeks to “prohibit distribution of computer-generated obscene material depicting a child” that is created through use of AI. This bill was designed to address concerns of AI scraping the faces and other biometric data of minors from existing media to use in the generation of indecent content. Under this bill, it would not require that the “child depicted actually exists” for a crime to have occurred.
Both bills easily passed the House and Senate committees, but neither received a vote on the Senate floor.
But the bill that has received the most notoriety out of this group has been H.B. 986, which seeks to limit the growing problem of AI deepfakes being used to interfere with elections, as seen in the New Hampshire primaries.
H.B. 986: Addressing Deepfakes in Elections
In its first incarnation, H.B. 986 was a simple short 3-page bill.
“This is a beginning of a conversation,” Thomas said during the bill’s first hearing, a joint meeting of the House Technology & Infrastructure Innovation and Judiciary Committees. And it most certainly was. The bill would see three more lengthy hearings, plenty of questions from committee members, and numerous versions before passing out of its first committee. Detailed and substantive conversations were had on various relevant issues, such as legal venue and means by which those affected by a deepfake could seek relief in court. By the time the dust had settled and the bill was ready to proceed to the full House, it was an 8-page bill.
The result was this: a bill that makes it a crime for an “employee, staff member, or other similarly affiliated individual of a political party, political body, candidate, campaign committee, political action committee, or any other political committee” to publish, broadcast, stream, or upload “materially deceptive media” with knowledge of its falsity “within 90 days of an election with the intent to deceive one or more electors for the purpose of… significantly influencing a candidate’s chance of being elected in such election… creating confusion about the administration of such election; or… otherwise influencing the result of such referendum.”
“Fraud is not protected free speech, and that’s what this bill is going after”
In short, deceptive and malicious deepfakes like the Biden robocall in New Hampshire could not be used in campaigning. But, that does not mean that campaign use of AI, or even deepfakes, would be completely banned under this bill. The legislation provides for the use of a disclaimer that can be used for AI-generated content, clearly stating who sponsored the content, who is being impersonated, that “events depicted may not have actually occurred,” and that “artificial intelligence was used to make part or all of the content of the campaign advertisement.” With that disclaimer, content could not be said to be “deceptive” and would not fall within the purview of this bill.
“Fraud” versus “Free Speech”
As H.B. 986 stood before the full House of Representatives, the bill enjoyed rather broad bipartisan support. But the bill was not without its detractors.
On the day before the vote, Mallory Staples, the Georgia Director of the State Freedom Caucus Network, wrote on her Substack, “We need to kill H.B. 986… The bill has a substitute and is less scary, but going from Stalin to Gorbachev is still communism… Infringing on free speech, regardless of how many parameters you put on it, is a HARD NO.”
During the floor debate on the bill, Representative Charlice Byrd (R-Woodstock), Chairwoman of the Georgia State Freedom Caucus Network, rose in opposition to the bill over free speech concerns. “This bill is an affront on our First Amendment rights,” she said. “We are gathered at a crossroads, faced with a bill that, while polished to a deceptive shine, threatens to erode the bedrock of freedom.”
From the House floor, Thomas spoke to the bill’s free speech safeguards. “We have put significant protections in this bill to ensure the judicial branch understands the will of the General Assembly and its dedication to the First Amendment.” Referencing language from the bill, he continued, “Nothing in this code section appl[ies] to activities protected by the First Amendment of the United States Constitution, including, but not limited to, satire, parody, or ‘memes,’ works of artistic expression, or works of journalism – and anybody that tells you otherwise is just flat out lying to you.”
“Fraud is not protected free speech, and that’s what this bill is going after,” Thomas said.
Speaking in favor of the bill, Chairman Todd Jones (R-South Forsyth) of the Technology & Infrastructure Innovation Committee, pointed to the unique problem of misinformation as posed by generative AI, and addressed why this bill is necessary. “It used to be that the lies that went through the political process… were twodimensional,” he said, “but now we’re talking about three-dimensional in terms of video, in terms of audio.”
The bill ultimately passed the House with a vote of 148-22.
In a hearing before the Senate Judiciary Committee, the bill continued to make headlines, due in part to a striking presentation of the powers of generative AI. The committee viewed two videos that drew stark contrast between the crude generative AI technology of merely a year ago and the sophistication of current technology. The most remarkable moment of the hearing, however, was a video presentation of the legislation — as offered by AI-generated audio testimony of two of the bill’s staunchest opponents: Senator Colton Moore (R-Trenton) and Mallory Staples. “I want all of you to know that I was wrong about H.B. 986,” the AI impersonation of Staples continued, “this bill is vital to protecting the integrity of our elections.” Amongst audio and visual disclaimers that the content was AI-generated and may not have actually occurred as prescribed by the legislation, the AI depiction of Moore stated, “H.B. 986 is a great bill that addresses one of the single most imminent threats to our election integrity.”
The video made waves in the committee room and beyond. “I think it was kind of [a] surprise to the Senators that sit on that committee [to witness] how easy it is to actually make a deepfake,” Thomas said regarding the video.
The bill subsequently passed out of the Senate committee with just one dissenting vote but never received a vote in the full chamber.
The Future of AI Legislation
AI continues to permeate everyday life, and states will undoubtedly continue to address concerns related to the technology, particularly as they become relevant. Such was the case with the elections deepfake bill. As the power of generative AI has become clearer and its uses as a tool of political deception have been witnessed in real-world scenarios, policymakers feel that it is time to pass legislation on the issue.
“Many people believe, including myself, that AI is a really powerful tool and it has some fantastic, exciting uses going forward,” Thomas says, “but we felt that with the 2024 election cycle going into full swing, we needed to immediately address some of its potential nefarious uses.”
Certainly, the AI regulatory landscape has its challenges, and legislation on the issue often faces a difficult path. However, it seems that there is a sustained interest among state lawmakers to address this problem. As technology development continues, so will important policy conversations. It seems to be that AI is here to stay — and so is AI legislation.
