Pennsylvania’s House passed a bill to bar AI deepfakes in election campaigns. Will it be enough to keep voters from getting fooled?

Pennsylvania’s House of Representatives has unanimously passed a bill regulating the use of artificial intelligence “deepfakes” in election campaigns, joining the growing list of states trying to put guardrails on the technology and how it’s used.

“It’s becoming so hard to tell what is real and what’s not real,” said State Rep. Tarik Khan, a Democrat who is the bill’s prime sponsor, whose district covers East Falls, Roxborough, and other northwest Philly neighborhoods.

“We have to make sure that our regulations are keeping up with the technology to protect the public,” he said.

“Deepfakes,” or hyperrealistic generated videos, pictures and audio usually created with the intent to deceive, have become increasingly widespread and easier to make as AI technology has improved. More than two dozen states have passed laws regulating AI deepfakes in elections, and now Pennsylvania joins 13 others with pending legislation.

Deepfake campaign ads have been used across local, state, and national races in recent years, including this spring in New Jersey’s gubernatorial primary. The former mayor of Shreveport, La., said a deepfake advertisement depicting him as a failing high school student was “100%” part of why he lost reelection in 2022.

Last year in North Carolina, a Democratic donor made ads with the AI-generated voice of Republican candidate Mark Robinson uttering real, outlandish quotes that he made in other venues, like that “the Civil Rights Movement was crap.” The best-known case so far has been a deepfake robocall impersonating Joe Biden that encouraged New Hampshire residents not to vote ahead of the 2024 presidential primary. The man behind the calls was acquitted of the most serious voter-suppression charges afterward, though the telecom company that broadcast the message agreed to pay a $1 million fine.

Though most deepfakes so far haven’t had clear race-altering impacts, and campaigns have generally resisted using AI ads, that could change as the technology is refined and its use becomes even more common.

Should it be enacted into law in Pennsylvania, House Bill 811 would require campaign advertisements disseminated within 90 days of an election that use deepfakes to clearly disclose that what is being shown did not actually occur.

Without this disclosure that an ad is using a deepfake, its disseminator may be subject to civil penalties and daily fines. In a race for a municipal public office, a court may issue fines up to $15,000 per day the advertisement is shared, whereas in a presidential or congressional race, fines may reach $250,000 per day.

The bill now moves to the state Senate for consideration.

To this point, AI’s development and growing use has outpaced many efforts to regulate it. That includes the many unresolved legal questions in how and to what extent students should be allowed to use AI in schools, or how liable AI companies should be when people develop intense emotional relationships with chatbots.

Meanwhile, the realistic nature of deepfake images and videos have vastly improved. It used to be easier to spot mouths that didn’t move quite right, stilted speech or extra fingers, but now, the delineation between real and fake is more easily blurred.

Creators mostly generate this media for purposes outside of influencing elections, like farming social media engagement or making their favorite celebrities look good.

Legislators have slowly made some progress on the basics. Gov. Josh Shapiro signed a separate bill into law on July 7 that amends a state statute to include deepfakes as a form of digital forgery. In October, Shapiro approved another deepfake bill, which prohibits the use of AI to create child pornography and non-consenting adult deepfake images.

The path is clear for more guardrails to go up, after the U.S. Senate voted by 99-1 to strike a proposed 10-year moratorium on state regulation of AI from President Donald Trump’s spending bill this month (though any state AI bills signed into law before the federal spending bill took effect would have been able to stand).

Khan said that he and the Pa. bill’s other sponsors, including lead cosponsor Republican Rep. Jeremy Shaffer of Allegheny County, decided to focus on elections because it’s the subject they know better than anyone and that maintaining the integrity of elections was crucial as technology speeds forward.

The public’s confidence in Pennsylvania’s elections has been particularly volatile both after the 2020 presidential race and leading up to 2024 elections. Though there was no widespread fraud in either case, Trump and his supporters made numerous accusations that the 2020 election was stolen in Pennsylvania, dominating social media, and being a catalyst for the Jan. 6 insurrection.

Khan’s legislative team took inspiration from other states’ AI election laws as they wrote the bill, and sought to find the space between establishing safeguards for the public while maintaining rights to free speech, while not stifling creativity.

State Rep. Paul Takac of Centre County, a Democrat who is a member of the House’s Communications and Technology Committee, said that while the bill is a good first step, he doesn’t think it goes quite far enough.

“There’s a lot of opportunities for bad actors to undermine our democratic process. That window of opportunity does not just exist 90 days prior to an election, and it doesn’t end at midnight on election night,” he said.

Takac supports the bill’s passage, but would like to strengthen the regulation in the future. “I think we need to be open to expanding these restrictions,” he said, suggesting that the law be broadened to also cover people who aren’t directly affiliated with campaigns who make malicious election deepfakes, as an example.

Though the bill is not aggressive, it is still important that Pennsylvania begins addressing AI deepfake use in elections, said Mekela Panditharatne, senior counsel for the Brennan Center for Justice, a nonpartisan law and policy institute.

“I think it is vital that states take action to address this issue,” she said, describing the Pennsylvania bill as being fairly similar to other states’ election deepfake laws. She believes that part of the reason that there has not yet been an onslaught of vicious deepfake campaign ads is that the pieces of state legislation passed across the county have acted as a deterrent.

“It seems to be the case that responsible campaigns are careful about using generative AI. And I don’t know that that would’ve been the case without the conversation that happened around it,” she said.

She agreed there is a point where more warnings of AI abuse may shift people’s perception too far in the opposite direction, where they begin to distrust reality. Called “The Liar’s Dividend,” the explosion of public knowledge of deepfakes means that people can more persuasively lie about the truth being AI-generated, like when Trump claimed that images of a large Kamala Harris campaign rally crowd were fake.

“A higher level of skepticism towards digital content is required in the generative AI era than was previously required, but that can also have downsides when authentic information is more easily questioned or distrusted,” Panditharatne said. “So public education efforts need to be balanced and prudent and not make people reflexively question content particularly when it could affect their right to vote or some other important consideration.“

Khan is excited about the promise of AI, and hopes that the technology continues to be improved and innovated. But with such explosive potential, he said protections have to guide its rise.

“We went from using the abacus to computers. It’s that big of a leap,” he said.

The bill now goes to the state Senate’s Technology and Communications Committee for consideration. The Senate is not slated to be back in session until September.