Michigan targets ‘deepfakes,’ campaign ads that use artificial intelligence

• Michigan lawmakers want to require disclaimers on political ‘deepfakes’ and campaign ads that use artificial intelligence
• Bills would establish fines and criminal penalties for creators or distributors who fail to include disclaimers
• Legislation is before the full Michigan House after Tuesday approval by the House Elections Committee

LANSING — Michigan state Rep. Penelope Tsernoglou broke into a wry smile on Tuesday as she played a recording of President Joe Biden appearing to praise her new legislation.

"No more malarky," Biden was heard to say about Tsernoglu’s bills, which  require disclaimers on political "deepfakes" and campaign advertisements generated using artificial intelligence technology.

The East Lansing Democrat then offered a disclaimer of her own: The recording she played for colleagues Tuesday in the House Election Committee was fake audio that a friend made in about three minutes using an artificial intelligence tool.

"AI-generated content is currently indistinguishable from real-life images and sounds," warned Tsernoglou, who chairs the House panel. "The threat of AI-generated content to influence and deeply impact elections and voters is imminent."

With online tools from OpenAI and other tech firms already bringing artificial intelligence to the masses, Tsernoglou and bipartisan co-sponsors are proposing new laws that would require disclaimers on a potential wave of campaign ads and other "materially deceptive" political content.

That includes so-called deepfakes: Content that is digitally altered in an increasingly convincing fashion to spread false information. 

Under legislation approved by the committee Tuesday and now headed to the House floor for further consideration, failure to include disclaimers could lead to criminal penalties for anyone, whether they’re campaign operatives or average citizens trying to support their candidate. 

Related:

• Bills adding ‘Obamacare’ provisions to Michigan law headed to governor
• Michigan ranks third for voter registration. Democrats push for more
• Lawsuit aims to disqualify Trump from Michigan ballot

Penalties would be stiffer for political professionals who could face up to 93 days in jail for a first offense, and longer for subsequent offenses. Average citizens would initially be subject to a $250 fine, which Tsnergolou called a warning to deter additional violations that could lead to jail time. 

Any individual that repeatedly produces deepfake videos or other intentionally deceptive media within 90 days of an election could eventually face up to five years in prison. 

Four states have already adopted similar legislation — Texas, Washington, California and Minnesota — and the Federal Election Commission is considering regulations to crack down on “deliberately deceptive artificial intelligence campaign ads.”

While rare, artificial intelligence has already been used in political content, including a social media video distributed by presidential candidate Ron DeSantis' campaign in June that falsely depicted former President Donald Trump hugging Dr. Anthony Fauci. 

In another instance, a DeSantis super PAC ran a TV ad that used an AI-generated audio version of Trump attacking Republican Gov. Kim Reynolds of Iowa, a state that will hold the nation's first caucus and play an important role in deciding next year's Republican presidential nomination. 

"Although deepfakes are not yet prevalent in American politics, they have been used, and I think it's a near certainty that they will become prevalent absent intervention from legislators in Michigan and around the country," said Robert Weissman of Public Citizen, a national democratic advocacy group that supports the new regulations. 

The Michigan bills would require political deepfakes or other "materially deceptive media" created, published or distributed in the state within 90 days of an election — including on social media — to include a disclaimer that they contain images, audio or video that "has been manipulated by technical means and depicts speech or conduct that did not occur.” 

Television, radio or print political ads paid for by campaigns or ballot committees would have to include similar disclaimers if they are created in whole or part using artificial intelligence, which would be defined as "a machine-based system that can make predictions, recommendations, or decisions influencing real or virtual environments for a given set of human-defined objectives." 

While the package is co-sponsored by GOP Rep. Matthew Bierlein of Vassar, the two Republicans on the House Elections Committee “passed” on each bill Tuesday rather than vote on them and questioned the rush. 

The bills were introduced last week and “I just think that there is a lot of things that could be fleshed out still,” said Rep. Rachelle Smit, R-Martin. “AI is here, but we definitely need to take some time to work on the measures and maybe hear a little bit more from the experts.”

Democrats on the committee supported the bills despite some concerns, particularly over potential criminal penalties.

Rep. Jamie Churches, a Wyandotte Democrat and former teacher, said she feared young people “may not understand the gravity of this” and could be so focused on going “viral” that they won’t consider the political or criminal implications. 

Likewise, Democratic state Rep. Dylan Wegela of Garden City said he wants to see a clear distinction between “uncle Bob creating their own deep fake” and “political actors intentionally misleading in elections.” 

Tsernoglou, the committee chair and lead sponsor, said she is “open to improvements” on the legislation and invited colleagues to draft potential amendments as the bills head to the House floor. 

But it is important to move quickly, she said, because the Legislature will need to finalize the plan in the next two months if Michigan is going to begin regulating AI-generated political ads in time for next year’s campaign season. 

“We need to keep moving forward if we want that to happen,” Tsernoglu told reporters after the hearing. “AI is here. Deepfakes are here. They’re happening now, and we need to address them.”