WHQR's Sunday Edition is a free weekly newsletter delivered every Sunday morning. You can sign up for Sunday Edition here.
Every day, for the last few years, I’ve woken up and thought, ‘When will someone here in the Wilmington area finally combine my two favorite things: negative campaign ads and AI slop?’
Well, just recently, my prayers were answered.
Two weeks ago, someone sent me a link to a new campaign ad, released by Michael Lee’s committee, featuring AI-generated imagery of Jessica Bichler, his Democratic opponent in the race for State Senate District 7. Bichler, who owns Wilmington Yoga, is portrayed leading a session with two other practitioners as they do some basic stretches, setting up the ad’s putative punchline.
“Some liberals lean to the left, some stretch all the way to woke, but yoga instructor Jessica Bichler, she's totally bonkers,” a voiceover intones.
The deepfake ad goes on to make two claims of leftist extremism — more on that, in a minute — while the video shows AI-Bichler in a cartoonish contortion, one leg curled like a swizzle straw around her arm, and swaying with an expression that, I assume, was intended to convey ‘bonkers.’
The ad also includes a cheeky disclaimer: "You guessed it! AI was definitely used to generate these silly video clips."
Pretty quickly, I started getting other messages about the video — some amused, some outraged, some who felt it was, all other things aside, a novel campaign move, at least in our local burg.
“You should thank Lee,” one listener wrote to me. “He’s given you next week’s column.”
Indeed — although I pushed this column off a week because Lee wasn’t available, and I thought it was important to get him and Bichler on the record about this ad, the political claims it makes, and more so, how the use of AI lands personally, politically, and legally.
I also wanted to take some time to check my initial — and somewhat illiberal — reaction to the AI ad.
In the off chance you missed the sarcasm above, I’m very much an AI skeptic, and loathe the ‘AI Slop’ of quasi-real music, art, and video that’s flooding the internet. (“Slop,” truly, is the perfect word for it.) And because of that, my initial reaction to the ad was, ‘surely this should be illegal.’ I’m clearly not alone, because some states, including North Carolina, are considering or have passed laws regulating and even banning deepfakes.
But I wondered, can that be squared with the First Amendment? Are some attempting to ban it because it's truly, categorically different from other kinds of protected speech — or because we’re reacting emotionally to it? I called up the folks at the Foundation for Individual Rights and Expression (FIRE), which is sort of like a Libertarian ACLU, to test out my gut reactions.
Nasty, brutish, and expensive
First, it’s worth noting that, even in the bloodsport of politics, the race for the state Senate seat representing most of New Hanover County (minus the gerrymandered ‘notch’ in downtown Wilmington), has been consistently nasty and expensive.
A decade ago, Lee defeated Democratic challenger Andrew Barnhill after a notably negative campaign that left both candidates frustrated. But in that case, the ads were paid for by the state GOP — and Lee told the StarNews that he didn’t authorize or approve of them, and said he actually asked the state party to stop.
Since then, it’s felt the race has grown meaner and more personal. In 2022, Lee and Democratic challenger Marcia Morgan burned through over $3 million, much of it spent on aggressive campaign ads. One of Morgan’s ads went too far, very likely into defamation territory, and Lee filed suit. Morgan was forced to admit one of her ads was “not based in actual fact,” later settling. Two years later, Democrat Dr. David Hill sent Lee a cease-and-desist letter based on his ads.
While both state parties continue to pour money into the race, candidates are no longer hiding behind Raleigh machines. Case in point: the AI ad is paid for by the Committee to Elect Michael Lee, not the NCGOP or the Senate Caucus.
To some, this probably feels like a top-down result of increasing rancor in DC, where rhetoric that would have been career-ending a generation ago is now par for the course (Trump is, of course, both a symptom and accelerant of the phenomenon, but he’s hardly alone). It fits with apocalyptic “the ceremony of innocence is drowned” vibes of Yeats’ “The Second Coming,” with its rough beast slouching towards Bethlehem (a mental image for the ages, if ever there was one).
But, with respect to the great Irishman, some would say it’s the same weather on a different day — some of our founding fathers were on both ends of Jerry Springer-level accusations (some of them verifiably true).
I think, in general, folks are less critical of a candidate’s attack ads when they’re ‘on the same team,’ but overall they’re still pretty unpopular. And it won’t shock you to learn that, like a majority of my fellow Americans, I dislike negative ads. To be honest with you, as a journalist, they’re a pain in the neck. The claims made by a 30-second ad, or even a door hanger, can take days of work to unpack and vet out. And it’s not just that attack ads tend to dwell in the murkiest borderlands of ‘truth'’ — it’s that even their legitimate claims are overshadowed by bombast and unsubtle suggestion.
The claims in Lee’s ad, for example, are a mixed bag: some kernels of truth, some things that seem false (or at least, unsubstantiated). It also struck some as just plain cruel. In addition to criticizing Bichler’s politics — which, at least on paper, is totally fair game — the ad also conspicuously tries to make her look physically ridiculous, and mocks the practice of yoga (including a shot of the AI-generated Bichler striking a gong, presumably for comedic effect). Some have dismissed this as trivial, saying it’s obvious parody. But others have told me they felt Lee was unfairly disparaging Bichler, not only a woman and a yoga instructor, but also a small business owner.
That’s a gamble candidates take when they go negative. Lee’s ad is unlikely to deter Republican voters from supporting him, and it’s even less likely to sway Democrats to his side. When it comes to the unaffiliated — who make up almost half of New Hanover County voters — who might be on the fence, a negative ad runs the risk of turning people off.
But it’s a risk politicians keep taking.
Experts have a simple explanation for the persistence of attack ads: they work. So, as journalists, I think we’re gonna be dealing with them for the foreseeable future — doubly so in the State Senate Race for District 7.
Fact-checking the ad’s claims
Before we get to the AI issues, I want to talk about the two claims Lee’s campaign makes in the ad. (If you’re just here for the deepfake of it all, you can jump ahead — I won’t judge you.)
The first is that Bichler’s campaign is “supported by groups who'd defund the police, let criminals walk free, and give free healthcare to illegals.” Like many similar ads, the video shows a brief reference for its claims, in this case, “American Friends Service Committee, 2/26.”
It’s hard to imagine what a layperson, who does not have both candidates' cell phone numbers, would do with that information. So, here’s what I could find.
American Friends Service Committee (AFSC) was founded by Quakers over a century ago, and currently operates as a nonprofit based out of Philadelphia. While the group might take issue with the ad’s phrasing, its strategic goals do include “the abolition of prisons, jails, and detention centers,” and offering “legal services, training, human rights monitoring, and humanitarian relief” to support migrants.
When I initially asked Lee about it, he provided a screenshot of what appeared to be a web search result, showing the name “American Friends Service Committee” above a link to NGP VAN, a data platform used by Democratic campaigns, and a link to a launch party for Bichler at Wrightsville Beach Brewery on January 27. As for “2/26” — that appears to be the date of the search.
Bichler told me the claim was completely false, and that she suspected it was due to a glitch in the search results, and — as first reported by Port City Daily — has pushed Lee to retract the claim.
“I had never even heard of that group until I saw it in Michael Lee’s deepfake ad, so yes, a cease-and-desist was sent to address those completely unfounded claims,” she said.
Last week, Lee told Port City Daily his team was “looking into it.” As of earlier this week, Bichler said she hadn’t received a response. In two different emails, I asked Lee if he had any additional information to support the link between AFSC and Bichler, but he didn’t address the issue in his response beyond sharing the screenshot of the search result again.
I tried repeatedly to contact AFSC’s national and North Carolina offices, but did not receive a response. Notably, as a 501(c)(3), AFSC would be prohibited from endorsing or opposing a political candidate — so Lee’s claim is likely to be an issue for them.
At best, I would say this claim is flimsy, and I think it is probably false. If I were working on oppo research, I would have kept digging before I ran with this one. I couldn’t find any other record of interaction between Bichler and AFSC, including in her promotional materials and campaign finance — and Lee hasn’t offered any additional evidence, although in his statement sent on Friday, he said in general his ad “didn’t deceive anyone.”
For the record, I asked Bichler if she’d weigh in on her actual policy.
“I’ve made my positions very clear: I believe in fully funding local law enforcement and making sure that our justice system has the resources it needs to keep criminals behind bars,” she told me. She didn’t comment on immigration, but said in general, “the claims made in Michael Lee’s ads trying to attach me to anything else are just as truthful as the deepfake he made — completely false.”
The ad’s second claim is that Bichler was “recruited to run by socialists,” with a reference to “Strict Scrutiny, 2/16/26.”
Ok, so you have to do a little work for this one.
Strict Scrutiny is a podcast focusing on the Supreme Court and legal culture, created in 2019 and acquired by Crooked Media in 2022. Crooked Media is a progressive-left media company, founded in 2017 by former aides and speechwriters for Barack Obama, and is probably best known for its flagship podcast Pod Save America.
In 2018, Crooked Media launched Vote Save America to support progressive causes and candidates. Vote Save America has since become an independent PAC, but it’s still closely associated with Crooked Media.
During the recurring Oprah-style ‘favorite things’ segment of the February 16 edition of Strict Scrutiny, co-host Kate Shaw gave Bichler a shout-out and asked listeners to support her. (The episodes have transcripts; I don’t think Lee’s team spent hours listening to the show, trawling for material.)
“Jessica, we were so excited to learn that you’re a Strict Scrutiny listener and that the show played a part in your decision to run for the State Senate,” Shaw said. “Jessica heard about Vote Save America’s candidate recruitment program on our show and filled out an online form, not knowing where it would lead and is now running for North Carolina State Senate District 7.”
According to campaign finance records, Vote Save America has given $6,800 to Bichler’s committee, the maximum allowable contribution.
So, “recruited,” in the sense that Vote Save America (and Crooked Media) works to get progressive-minded folks to run for office and support them, I think can hold some weight. Although, as I understand it, the process also involves the Democratic State Senate Caucus. It’s a lot of work to put together a political campaign, especially for a big state-level race; it takes a lot more than filling out an online form.
And “socialists” — well, that depends on whom you’re talking to. I asked Lee if he was using the word in a casual sense, the way some conservatives refer to much that's left of center as ‘socialism’ (or, increasingly, ‘communism’) or if he was claiming more specifically that Vote Save America (and thus by extension Bichler) was affiliated with the Democratic Socialists of America. He didn’t answer that question.
For a middle-of-the-road voter, watching the ad casually, “recruited by socialist” might inflame the kind of concerns a lot of centrist and center-left have about more radical left-wing politics. But once you unpack it, it’s a bit of a stretch. Actual socialists, of the DSA variety, would probably dislike the kind of pragmatic, big-tent incrementalism that undergirds a lot of Pod Save America conversations (which sometimes criticize them). But you’re not going to get that kind of political nuance from a campaign ad — and, to be fair, there are center-right politicians who will, no doubt, be labeled ‘far right’ or ‘right wing’ when that’s not really accurate, either.
A slop in the face
But, of course, this was not just an attack ad, it was an AI-generated deepfake attack ad.
For a lot of people I’ve spoken to about this, there is something different about this ad compared to the ones we’ve grown used to seeing. That is, I suspect, part of the point. People dislike attack ads, but many have learned to tune them out. Some found this ad concerning or cruel, others found it silly or stupid, and a few actually thought it was funny. But I think they’d agree: it was harder to ignore.
In terms of using AI in general, Bichler criticized Lee for making “a deliberate choice to support data centers powering AI use over supporting our local film community. Instead of employing a highly skilled local film crew, he decided to buy AI slop.”
I doubt Lee’s ad would have been any more well received by his detractors if he’d gone with a dramatic reenactment over a deepfake. But there are certainly things about the AI industry — the theft of intellectual property, the environmental concerns, the looming job losses — that rub people the wrong way, and using it in a campaign ad is certainly a risk for that reason.
When it came to Lee’s ad in particular, Bichler said she was disturbed.
“I first noticed the AI in the mailers, where I honestly wasn't sure the person pictured was supposed to be me. The TV ad is something else entirely. Seeing my likeness moving around like that on screen was disconcerting. The idea that this man not only has the power to create these images, but feels comfortable making my body and my voice do absolutely anything that he chooses is incredibly creepy. The reaction I'm hearing, from people of many political leanings, is that it’s deeply wrong and frankly, weird,” she said.
There are lots of ways to misrepresent people using real sound and images. You can use an unflattering photo, or an awkward soundbite. You can take comments out of context, or use traditional editing techniques to make someone look unattractive or appear to react inappropriately to an event (laughing at a memorial, or looking dour at a celebration). The opponent often appears in greyscale, like a crime show reenactment, with ominous music playing, while the candidate appears in warm, bright color, with a stirring soundtrack.
Bichler identified something a lot of people I’ve spoken to also felt: the creation of imagery that didn’t previously exist seems categorically different than simply manipulating or misrepresenting a photo or video.
Lee acknowledged that he approved this ad, so it’s fair to say he thought it was fair game. But I asked whether he thought there were — or could be — cases where it went too far.
“There is a line, and it isn't about whether a computer was involved. The line is deception. No one should use AI to make a voter believe a real person said something they never said or did something they never did,” Lee wrote in response. “This ad does not deceive anyone. The image of my opponent is clearly a caricature — no human body can be twisted like that. And she doesn't speak in it. Nobody would believe it’s a real event.”
It’s a fair point that no one is likely to have seen Lee’s ad and thought it was real. But at the same time, AI technology to render realistic images has developed rapidly. Just look back two years to the AI-generated ad attacking Republican Mark Robinson, who was then running for Governor. The graphics are horrible, although the ad itself was pretty effective, relying on the inflammatory and probably disqualifying things Robinson had actually said. Point being, at the time, I don’t recall anyone, on either side of the aisle, fretting about the ethics or morality of using Robinson’s likeness.
Lee argues that the fakeness of this deepfake is so conspicuous that, even without a disclosure notice (which the ad does include), no one is at risk of believing it’s real. And yet, the fidelity with which it recreates Bichler’s likeness is significantly better than any cartoon or doctored video.
So, I think it’s fair to consider what happens with the next ad, and the one after that, with imagery — and speech — that does appear real.
Regulating AI campaign ads
State legislators in North Carolina, including Lee, have started addressing the harmful potential of deepfakes and AI-generated campaign material. Three years ago, Lee co-sponsored Senate Bill 88, which would have required disclosures on any campaign ad that featured AI-generated content.
Though the bill died in the House, Lee told me when it came to his own campaign ad, he “still added a disclosure because it’s the right thing to do.”
He also said he’d be in favor of regulating “deceptive” deepfakes, which wouldn’t exclude the way he used AI in his ad.
“North Carolina needs a law to regulate the deception, not the tool. These tools are here, and they are not going away. The obligation to use them honestly sits with whoever uses them. That includes me. And that’s exactly what I did,” Lee wrote.
Bichler took a stronger line against the use of deepfakes.
“Asking for honesty about how a message was made isn’t a restriction on free speech. Opposing the theft of a person’s likeness to build a fantasy about her isn’t censorship, it’s a defense of basic dignity. If you have something to say about me, you should have the courage to say it without hiding behind a deepfake,” she wrote.
Earlier this year, a bipartisan group of state representatives in the House filed a bill banning deceptive deepfakes that depict “a natural person speaking or acting in a manner that the person did not actually speak or act” to harass or harm someone, or with the intent of “injuring a candidate for elected office or influencing an election.” That would likely include Lee’s ad, or similar deepfakes.
The bill, which has been parked in the House Elections Law Committee since May, would make the creation of an illegal deepfake a Class 1 misdemeanor (the second most serious type).
Overall, 31 states have passed laws regulating AI-generated content, including deepfakes, in political messaging, according to the National Conference of State Legislatures.
Most of those states focused on disclosure requirements, with 28 requiring the same sort of notice you’d see identifying who paid for a political ad. Colorado and Utah also require disclosure in the metadata of video and images.
Three states go further, prohibiting certain AI use. Minnesota and Texas prohibit deepfakes in a specific time period prior to the election. Maryland’s law bans election-related deepfakes in general, with no time period.
AI and free speech
Again, I’m no fan of AI. I dislike it for the same reasons a lot of people do: it’s a cheap facsimile of human creation that’s boiling lakes’ worth of water and setting the scene for catastrophic job losses. It also poses serious threats as a tool of misinformation and propaganda. The 2022 deepfake of Ukrainian President Volodymyr Zelensky ‘surrendering’ to Russia was mocked as amateurishly unconvincing, but the technology is getting better and more deceptive.
There are a lot of reasons to be concerned. But, at the same time, I’m also a journalist, a civil libertarian by nature and in practice, and when governments get an itchy trigger finger around free speech, I worry.
I wanted to hash out my conflicting instincts here, so I got in touch with John Coleman, who serves as legislative counsel on AI and free expression for FIRE. We’ve called on FIRE before, on a host of issues, from New Hanover County Schools’ flag and display policy to CFCC’s attempt to censure a student performance.
Coleman told me FIRE strongly objects to laws seeking to ban deepfakes.
“I think you can dislike a political ad made with AI without wanting the government to come in and regulate it,” he said. “So, the First Amendment protects political speech, which means that the government has little power over what voters are allowed to see or hear during an election cycle. And that includes speech that is false, exaggerated, offensive, or even in poor taste, right?”
Many of the bills, Coleman argued, are overly broad.
“They’re capturing parody and satire, memes, and other forms of political speech, like normal commentary, and all of this is being exchanged, distributed, shared across social media,” he said. “And oftentimes, these bills are not just candidate against candidate, but someone sharing that kind of content online. So [the laws] are going after you, me, and Grandma when it comes to political speech.”
That objection to government intervention, for Coleman, also extended to laws requiring disclosure of AI-generated content. He argued that the would represent compelled speech.
There are, of course, limits to free speech – and Coleman noted that some of the potential misuses of AI would fall under defamation laws.
I conceded that, when it came to free speech, I would support something like the editorial cartoons in The Pender-Topsail Post & Voice, which are often mock and criticize some of the county’s political figures; regardless of how I felt about them, they’re protected speech. The likenesses of public figures are fair game for art, including satire, and journalism, including critical commentary. Viewed as just another tool, AI creates just another cartoon — even if it's a disturbingly realistic one.
Talking to Coleman, I tried to walk through a somewhat inchoate idea: that there was some way in which AI’s ability to uncannily create imagery of someone was not just superficially but categorically different from other methods. It’s been a long time since my cognitive psychology days, but I wondered out loud if, at a neurological level, a realistic AI image is processed differently than a cartoon or a cardboard cut-out. And, if that were the case, something about an AI video, even a satirical one, that wasn’t technically defamatory was still somehow different, and maybe outside the bounds of, other protected speech.
Not surprisingly, Coleman didn’t go for that, summing it up as “de facto defamation” just by using AI at all, though he was sympathetic to the concern. He noted that historically, new technology — the printing press, photography, radio — made it possible to spread all manner of ‘bad speech’ in faster and harder-to-check ways. That made sense to me, though — if I’m honest — I still can’t quite shake my sense that AI is a new beast, somehow.
Ultimately, Coleman offered the classic tonic of free speech advocates: the answer to bad speech is more, better speech, not censorship.
“What happens if you regulate this is, someone's going to get a cause of action, or some government official is going to be able to haul someone into court over a violation, which often means that the content is taken down or is temporarily blocked and litigated over in a very short time frame, all the while these people are running a race to to be elected,” Coleman said. “The most appropriate place to have these conversations is always in the public square. Allow for political debate to happen, and allow for more speech to combat false statements, and let that process work.”
The cure for bad speech
I’m not sure I’m fully convinced of FIRE’s point of view, although it was sobering to consider how quickly I mentally jumped to government intervention. I have to remind myself of the things that are protected by the First Amendment – yes, hate speech, but also conspiracy theories, misinformation, and, in many cases, outright lies. Separating my discomfort about that from the importance of free speech is not always easy. It’s not in this case.
I think it’s notable that Bichler and Lee both think disclosing the use of AI is not just allowable but important — and I agree with them. I think there’s potential for material harm that isn’t abated by defamation laws and, while we figure it out, disclosure laws make sense to me. The ones already on the books may well be challenged, and we’ll see how that goes. Maybe in ten years we’ll need them less, maybe we’ll need them more.
When it comes to banning AI-generated content, I think I’m closer to agreeing with Coleman and FIRE. For all the words I could use to describe AI — tacky, cheap, stupid, cringe — if it’s speech, I think it’s worth being careful about how we regulate it. Outside of the world of political messaging, where it involves people who aren’t public figures, I think there’s a different conversation when it comes to harassment and threats. (But for now, since we’re talking about political ads, I’ll leave it there.)
One thing I think Coleman and I agreed on is that — whatever you think of AI content — there is also some responsibility on the part of the individual, in this case, the voter.
Imagine a world with no regulation and no mandatory disclosure, where an extremely realistic AI-created video, virtually indistinguishable from real imagery, with no tell-tale quirks, is published, showing a candidate doing and saying reprehensible things. (In the very near future, you may not need to imagine it at all). What, as a voter, would you do?
One hopes that, at minimum, you’d search online for news coverage of the candidate by a respected outlet — and not just read the AI summary or the headline, but actually click on the link and check out the article. The antidote to bad speech is good speech, for example, solid, trustworthy reporting.
Polling shows that Americans feel negative campaign ads discourage them from civic engagement, even from voting, but those people could make a different choice. Instead of pulling away, they could dig in.
That choice is independent of what the ads look like or what new technology is used to create or distribute them. Fact-checking the actual claims in a state-of-the-art deepfake, at the end of the day, is pretty much the same as fact-checking something as low-tech as a door-hanger or mailer. A pain in the neck, but worth it.
As Jefferson put it, “Whenever the people are well informed, they can be trusted with their own government.” And when they’re not, they may be borne away on a tide of slop.
