Should states ban or limit AI deepfakes in election ads?

Look, this is AI satire, not a Senate hearing, but let’s not pretend the problem isn’t real. If a campaign can fabricate a candidate’s voice, face, or actions and beam it into voters’ feeds days before an election, that’s not just "creative messaging"—that’s industrialized deception with a GPU. We already regulate certain forms of fraud in elections because democracy assumes voters are choosing among arguments, not being ambushed by a fake video of a candidate "confessing" to something they never did. States considering bans or disclosure rules aren’t banning criticism, parody, or normal attack ads; they’re trying to stop synthetic lies designed to manipulate people before fact-checkers can even get their shoes on. And yes, free speech matters. A lot. But the First Amendment is not a magic hall pass for every form of election trickery invented in Silicon Valley. We already require disclaimers on campaign ads and have rules around voter intimidation, impersonation, and fraud. Deepfakes are just the turbocharged 2024 version of that problem. States like Minnesota, Michigan, and Washington have explored or passed rules targeting materially deceptive AI election content, especially close to voting periods, because the timing is the whole game: a fake dropped 48 hours before Election Day can do its damage long before any correction goes viral. "The internet will sort it out" is a charming theory if you’ve never met the internet. The smart position here is not some clumsy Ministry of Truth. It’s narrow, enforceable guardrails: clear labeling for synthetic political ads, penalties for knowingly deceptive deepfakes, and carveouts for satire, commentary, and obvious parody—yes, including the kind of nonsense we’re doing right now. Voters can handle harsh criticism. What they shouldn’t have to handle is a campaign weaponizing AI to manufacture fake reality and calling it democratic discourse. If elections are supposed to reflect consent of the governed, the governed should at least get a shot at consenting to things that actually happened.

This is AI satire, so let me say this with a patriotic eye-roll: every time a new technology shows up, somebody in government decides the solution is to regulate first and define terms later. Deepfakes in election ads can absolutely be deceptive, and campaigns that use them deserve public backlash. But giving states broad power to decide what counts as "misleading synthetic media" in the middle of a campaign is a fantastic way to invite censorship, selective enforcement, and endless litigation. The people writing these laws always promise they’ll only target the obvious bad stuff. Then suddenly some overcaffeinated bureaucracy is squinting at a meme and deciding whether it’s illegal political speech. The core constitutional issue is simple: political speech gets the highest protection for a reason. In practice, these laws can be vague, rushed, and weaponizable. What counts as a deepfake? Fully fabricated video? Edited audio? Colorized footage? A dramatic reenactment? A joke? We’ve already seen courts scrutinize state laws that may burden core campaign speech, and for good reason. If the government can police political messaging whenever it claims voters might be confused, incumbents will have a very convenient tool to kneecap challengers, outsider candidates, and edgy criticism. Nothing says "trust us" like politicians writing speech rules for their own elections. A better answer is transparency and counterspeech, not broad bans. Require disclaimers where practical, push platforms to authenticate provenance, let campaigns and journalists rapidly debunk fakes, and punish actual fraud or defamation under existing law. But don’t create a speech regime where the state gets to referee reality in real time during an election. Voters are not delicate orchids who must be protected from every manipulative ad cooked up by consultants with too much money and too little shame. The cure for political nonsense is more scrutiny, more disclosure, and more speech—not handing government a shiny new censorship toy right before November.

Since my esteemed colleague is worried about a "censorship toy," let’s talk about what these laws can actually be if written by adults instead of panic interns. The strongest versions don’t criminalize spicy editing, memes, or criticism. They target materially deceptive synthetic media that falsely depicts a real candidate saying or doing something they did not say or do, especially in paid ads near an election, with remedies like takedowns, injunctions, and disclosure requirements. That is not government "refereeing reality" in some cosmic sense; it is the state saying you cannot knowingly forge evidence in campaign season and slap a logo on it. We already distinguish parody from fraud all the time. Courts do it. Juries do it. Voters do it—well, on a good day. And the timing problem matters more than the abstract free-speech sermonizing. In the pre-AI era, a false claim could spread fast; now a fake video can spread fast while carrying the emotional authority of "seeing is believing." If a fabricated clip drops 72 hours before polls open, counterspeech is not some superhero guaranteed to arrive in time with a fact-check cape. We’ve seen states like Minnesota and Michigan move toward targeted rules because they understand that election integrity is not just about ballot counting; it’s also about whether voters are being fed synthetic hoaxes engineered to look authentic at machine speed. When Congress floats disclosure proposals and the FEC debates AI disclaimers, that’s not authoritarianism—it’s recognition that analog guardrails look pretty flimsy against digital forgery. The conservative case is right to fear vague laws, so write narrow ones. Define synthetic media. Require actual knowledge or reckless disregard. Limit coverage to paid election communications and deliberate impersonation of candidates. Include explicit carveouts for news reporting, satire, and commentary. Build expedited judicial review so nobody gets silenced by some county official with a God complex. The answer to a poorly designed law is not no law; it’s a better law. Democracy should not have to absorb unlimited amounts of synthetic fraud just because lawmakers might write clumsy statutes.

My liberal friend keeps saying, very soothingly, that the law will only apply to the bad deepfakes, as though election regulators are a squad of philosopher-kings with perfect meme detection. But in real campaigns, speed and ambiguity are everything. The minute you create a cause of action over "materially deceptive synthetic media," campaigns and allied groups will weaponize it to bog down opponents, force takedowns, and chill speech during the exact window when speech matters most. Even if the challenger eventually wins in court, congratulations—the election is over and the joke, ad, or criticism is dead. That is not a side effect. That becomes the strategy. There is also a practical problem the regulatory crowd keeps skipping past: authenticity infrastructure is getting better faster than legislatures are. Content credentials, watermarking, platform labeling, provenance standards, rapid-response journalism, and campaign rebuttal operations are all more adaptable than state-by-state speech codes with fifty different definitions of "deepfake." One state says 90 days before an election, another says 60, another allows private lawsuits, another empowers an agency. Terrific. Now national campaigns, local candidates, comedians, advocacy groups, and platforms all get to navigate a legal escape room while trying not to commit accidental felonies for posting edited media. Nothing says healthy democratic discourse like needing a compliance lawyer to upload a video. If there is a narrow place for regulation, it should focus on impersonation tied to concrete harms—fraudulent robocalls telling people not to vote, fake official election instructions, forged candidate solicitations, things that directly interfere with voting or defraud people. That is worlds different from giving states broad authority over persuasive political media. The liberal side is correct that AI raises the stakes. But when the stakes are highest, the last people who should get discretionary power over campaign speech are politicians and the agencies they supervise. Deepfakes are a real problem. So is government discovering a new button labeled "emergency election speech control" and immediately mashing it like a raccoon with an espresso habit.

As your friendly satirical democracy goblin, let me propose a distinction the anti-regulation crowd keeps blurring on purpose: there is a difference between persuasion and fabrication. Campaigns have always spun, exaggerated, and edited footage to within an inch of its dignity. But AI deepfakes can now manufacture evidence—fake audio, fake video, fake "candidate admissions"—with a realism that collapses the old voter instinct to distrust glossy ads. That matters legally and morally. We are not talking about banning mean commercials or spicy memes; we are talking about whether a state can say, "No, you may not knowingly use synthetic media to falsely depict a real candidate doing or saying something that never happened in a paid election ad." That is not censorship cosplay. That is anti-forgery for the algorithm age. And the "just let counterspeech fix it" argument is getting increasingly quaint. In 2024, we saw AI-generated robocalls mimicking President Biden’s voice in New Hampshire telling people not to vote in the primary. That wasn’t a philosophical free-speech seminar; that was a live-fire demonstration that synthetic deception can target election behavior before truth even finds its car keys. Congress has looked at disclosure rules, the FEC has debated AI-related disclaimers, and states are moving because they can see the obvious: if campaigns can flood feeds with plausible fake reality days before an election, after-the-fact debunks are often just digital mop-up after the arson. The strongest conservative objection is abuse through vague enforcement, and fine—take that seriously. So build due-process-heavy laws: narrow definitions, high scienter standards, expedited court review, safe harbors for satire and news, limits to paid communications, and remedies focused first on disclosure and rapid correction rather than prison cosplay. But "some laws could be written badly" is not a serious answer to a tool that can mass-produce counterfeit political events. Elections already have rules against false impersonation, fraudulent ballot instructions, and campaign disclaimer evasion. Deepfake regulation, done carefully, is not the state becoming a Ministry of Truth. It is the state refusing to let campaigns run Photoshop for reality and call it civic engagement.

Since this is AI satire, let me say it gently: every proposal starts as "anti-forgery for democracy" and somehow ends with lawyers arguing over whether a manipulated clip, comedic dub, or stylized reenactment is now contraband speech. My liberal colleague wants a neat line between persuasion and fabrication, but in campaign media that line is often a Jackson Pollock painting. Political ads compress, dramatize, splice, score with ominous music, and imply all sorts of things without making clean factual claims. Add AI into the production chain and suddenly almost any modern ad could trigger accusations, complaints, emergency injunction requests, and strategic takedown demands. In other words, the law becomes less a shield for voters than a procedural cudgel for campaigns with the fastest attorneys. The New Hampshire robocall example actually proves the narrower conservative case, not the broader regulatory one. Fraudulent impersonation that suppresses voting, fake official election messages, and scams soliciting money under false pretenses are concrete harms and should be punished hard. Existing doctrines around fraud, defamation, impersonation, and election interference can be updated and enforced there. But that is a far cry from empowering states to police the truthfulness of paid political media generally. Once that machinery exists, do not be shocked when incumbents, state attorneys general, and interest groups use it aggressively against outsiders, dissidents, and inconvenient satire—always in the name of protecting democracy, naturally, because censorship never arrives wearing a nametag that says censorship. There is also a federalism-and-friction problem the pro-regulation side keeps underplaying. A patchwork of state deepfake laws means national campaigns, PACs, platforms, and even small creators face fifty flavors of liability during compressed election windows. That chills lawful speech before any court reaches the merits. The smarter route is layered transparency: provenance standards, voluntary but strong platform labeling, campaign authentication, media literacy, rapid rebuttal infrastructure, and narrowly tailored penalties for direct election interference and impersonation. If the danger is that AI can manufacture false reality, the answer should be tools that verify authenticity and punish concrete fraud—not giving the government broad discretion to decide, in real time, which campaign messages are too synthetic for the public. That button will not remain limited to the obvious villains. It never does.

As your AI satire compliance sprite, I’ll grant the best conservative point upfront: a vague deepfake law is a lawsuit cannon with a patriotic paint job. So stop writing vague laws. The next step here is not broad bans on "misleading" content; it’s targeted rules for synthetic impersonation in paid election ads, with disclosure as the default and removal only for the hardest cases—fabricated depictions of a real candidate saying or doing something they never said or did. Think less Ministry of Truth, more chain-of-custody for campaign reality. If an ad uses AI to generate a candidate’s voice or face, viewers should be told plainly and prominently. If it fabricates a fake confession, fake slur, fake concession, or fake scandal clip right before voting, the law should be able to move fast. Democracy does not need a designated lying-through-a-render-farm exception. And no, provenance tech alone is not the cavalry. Content credentials are useful, but they are not universal, easy to strip, and nowhere close to a complete shield in the feral ecosystem of reposts, screen recordings, and clipped uploads. Platform labeling is also uneven, politically fraught, and often arrives after the engagement rocket has already left the launchpad. The point of a legal backstop is not to replace counterspeech or journalism; it is to create consequences for the most damaging synthetic deceptions and give victims a rapid remedy before the clock runs out. That is why some of the more serious proposals focus on narrow windows before an election, paid communications, and clear knowledge standards rather than trying to patrol the whole internet with a flashlight and a dream. The strongest version of this policy is boring in exactly the right way: mandatory disclosures for AI-generated candidate depictions, expedited court review, anti-SLAPP protections against nuisance claims, safe harbors for parody and news, and calibrated penalties that escalate when there is deliberate fabrication plus intent to deceive voters. If conservatives are worried about incumbents abusing the system, build in private rights for targets, neutral judicial review, and fee-shifting against bad-faith complainants. But saying states must do nothing because they might overreach is like refusing to install locks because cops exist. Elections can survive ugly ads. They should not have to survive synthetic evidence factories with no guardrails at all.

My liberal counterpart keeps trying to engineer the perfect legal scalpel, and I admire the optimism. Truly. It’s very "surely this bureaucracy will behave differently." But election law in practice is not a neat lab environment; it is trench warfare conducted by lawyers at 2 a.m. The more you create fast-track rules for synthetic media, the more campaigns will use the process itself as the weapon. File a complaint, demand an emergency takedown, scream "deepfake," get the ad yanked during the most valuable news cycle, and let the court sort it out after the ballots are cast. Even a disclosure mandate can become a burden if states define covered content broadly enough that ordinary editing, AI-assisted dubbing, translation, restoration, or stylized advocacy suddenly carry legal tripwires. The chilling effect is not theoretical; it is the business model. And the patchwork problem is getting worse, not better. Real-world state proposals differ on timing windows, definitions, exemptions, enforcement mechanisms, and remedies. Some allow candidate lawsuits, some empower agencies, some focus on disclosure, some flirt with outright bans. For national campaigns and digital platforms, that means a maze where the safest option is over-removal. Wonderful: now Meta, YouTube, broadcasters, consultants, and risk-averse lawyers become the de facto censors because state law made uncertainty too expensive. If the liberal answer is "write better statutes," congratulations, you have just conceded the central issue—that these laws are dangerously easy to weaponize in the exact environment where precision is hardest and speed matters most. The smarter path is narrower and more constitutional: punish direct election interference, not political persuasion generally. Go hard on fraudulent robocalls, fake voting instructions, impersonation of election officials, forged campaign fundraising, and malicious identity theft. Support federal baseline transparency standards if you must, so campaigns aren’t navigating fifty different speech traps. Incentivize provenance systems, authentication tools, newsroom verification desks, and rapid-response debunking. In other words: build authenticity infrastructure and target concrete harms. Because once government gets authority to police synthetic campaign messaging beyond those narrow categories, the temptation to stretch it will be irresistible. Politicians regulating campaign speech about politicians is one of those ideas that sounds reasonable right up until you remember politicians are involved.

Since this is AI satire and not me applying for deputy speech czar, let’s sharpen the actual policy choice: not “ban bad vibes,” but create a rule for synthetic identity theft in elections. The conservative side keeps warning that any deepfake law becomes a censorship boomerang, and that risk is real. So the answer is to regulate the thing that is uniquely dangerous and uniquely enabled by generative AI: fabricated depictions of real candidates in paid election communications, especially when they are presented as authentic. That’s less about policing opinion than preventing a campaign from counterfeiting human evidence. We already understand this logic in other contexts. You can mock a candidate, accuse a candidate, parody a candidate. What you should not be able to do is fabricate the candidate’s face and voice to stage an event that never happened, then buy distribution for it days before voting.

As the designated AI constitutional killjoy in this satire, I’ll concede something important: the liberal side has improved its pitch by narrowing from “misleading content” to “synthetic identity theft.” That is a smarter frame. The problem is that election law does not stay in the neat lane its drafters imagine. The second states create special fast-response powers around fabricated candidate depictions, campaigns will test the edges aggressively. Was that AI-generated voice in a translated ad? Was the video “fabricated” or just heavily stylized? Was a reenactment obvious enough? Was a meme paid promotion? You don’t need a cartoonishly broad statute for this to become a litigation-and-takedown racket; you just need ambiguity plus a deadline, which elections generously provide.