South Carolina AI Campaign Rules Lag as Political Content Gets Cheaper

South Carolina AI campaign rules remain unsettled despite generative AI becoming a significant force in the 2026 election cycle. Campaigns can now draft messages, analyze voter data, and produce digital media with fewer people and less money. That efficiency is widening access to professional campaign tools, but it also makes synthetic political content easier to create and harder to evaluate.

The conflict is no longer simply between campaigns that use AI and campaigns that do not. It is between inexpensive political expression and reliable signals about who created a message, what was altered, and whether a candidate actually said what voters see.

South Carolina lawmakers considered a targeted answer through legislation covering deceptive deepfakes. The proposal did not advance before the election season intensified. Meanwhile, 31 states had enacted political deepfake laws by June 2026, according to the National Conference of State Legislatures.

That leaves South Carolina campaigns, voters, newsrooms, and platforms working without a broad state framework designed for AI-generated political media. Existing laws still apply in certain situations, but they do not create a universal disclosure rule for synthetic campaign content.

This report distinguishes enacted law from pending proposals and voluntary practices. Legislative details were checked against the South Carolina General Assembly’s official record for H. 3517, including its status history and bill text. Federal claims are attributed to the Federal Election Commission, Federal Communications Commission, and U.S. Supreme Court. Examples involving individual campaigns are presented as reported cases rather than evidence that every campaign uses AI in the same way.

AI Has Moved From Campaign Experiment to Campaign Infrastructure

The immediate change is operational: AI now affects routine campaign work as well as the political advertisements voters see.

Generative AI, meaning software that produces new text, images, audio, or video from instructions, can shorten several campaign workflows. Staff members can use it to draft press releases, produce variations of social posts, summarize research, or create preliminary designs.

Gibbs Knotts, a political science professor at Coastal Carolina University, described that broad adoption in recent WBTV campaign reporting. He said AI was becoming part of campaign strategy, analysis, and press-release writing. His comments describe observed campaign practices, not a comprehensive measurement of AI use across every South Carolina race.

That matters because political technology rarely arrives in a single, easily regulated format. A campaign might use a language model privately to outline a speech. It might also publish an AI-generated image portraying an opponent at an event that never happened.

Those activities use related technology, but they create very different risks. The first resembles ordinary productivity software. The second can change what voters believe about a real person.

Campaigns also face pressure to produce more material for more channels. A single announcement can become an email, donation appeal, video script, image, flyer, and multiple social posts. AI reduces the labor required to make those variations.

That advantage is especially meaningful for local and down-ballot candidates. Smaller campaigns may lack dedicated researchers, designers, video editors, or communications consultants. An AI system can give them an inexpensive starting point for work that once required several specialists.

A Georgia legislative candidate offered a practical example during the 2024 election cycle. His campaign used AI-generated articles, fictional images, and a cloned version of his voice. The candidate told the Associated Press that the time savings allowed him to visit more voters.

The same case exposed the tradeoff. His opponent argued that AI-generated communication weakened the authenticity voters expected from candidates. The disagreement was not about whether the technology worked. It concerned what political communication should represent.

For a voter, the practical difference can be difficult to see. A campaign email polished with an AI writing assistant may look no different from one edited by a staff member. A cloned voice in a video can be much more consequential because the viewer may reasonably believe the candidate recorded the words. Without disclosure, the audience cannot reliably distinguish those workflows from the finished material alone.

A campaign worker would see another practical distinction. Using AI to turn an approved policy document into three draft social posts still leaves the original source available for comparison. Generating a realistic recording of an opponent saying words that person never spoke creates no authentic recording to check. The second workflow therefore requires disclosure and approval controls that ordinary drafting assistance may not.

This is why South Carolina political AI cannot be treated only as a deepfake problem. Much of its use is ordinary, invisible, and potentially beneficial. A campaign may use AI throughout its internal process without publishing deceptive media.

However, routine adoption builds the capacity to generate public content at scale. Once the same tools can produce realistic voices, faces, and events, the line between assistance and impersonation becomes politically important.

The key distinction is not whether a campaign used AI somewhere. It is whether voters receive the context needed to judge the resulting message.

Cheaper Production Helps Small Campaigns and Their Attackers

AI lowers the cost of participating in politics, but it also lowers the cost of flooding a race with misleading material.

Political campaigns have always depended on unequal resources. Well-funded candidates can hire consultants, commission polling, test advertisements, and respond rapidly to attacks. Local challengers often rely on volunteers and limited media coverage.

Generative AI narrows part of that operational gap. A candidate can use it to brainstorm messages, translate material, clean up audio, or create draft graphics. Human review remains essential, but the initial production burden falls.

That can improve political competition. A small campaign may communicate more consistently or redirect staff time toward meeting voters. The technology can also help people with limited design, writing, or video experience express ideas in more polished forms.

Yet accessibility works in both directions. The person creating a false clip needs fewer technical skills, less time, and a smaller budget than earlier forms of sophisticated manipulation required.

Deepfakes are synthetic images, recordings, or videos that realistically depict events or speech that did not occur. They are only one category of AI-assisted persuasion. Misleading content can also combine real footage, generated narration, selective editing, and false captions.

A fabricated video does not need to fool every viewer. It can succeed by reaching a small audience at a decisive moment, forcing a campaign to spend time denying it, or seeding uncertainty around authentic evidence.

The risk becomes sharper in lower-profile contests. National candidates attract reporters, fact-checkers, opposition researchers, and large communications teams. A county or state legislative candidate may have none of those defenses.

Associated Press reporting on down-ballot campaigns found that inexpensive AI tools can help candidates operate. It also found that smaller campaigns may struggle to counter AI-generated falsehoods.

The 2022 Shreveport mayoral race provided an early warning. A political committee used manipulated media targeting incumbent Mayor Adrian Perkins. According to the Associated Press, the content carried a label, but Perkins said his campaign lacked the resources to respond effectively.

In practice, a disclosure therefore does not automatically eliminate harm. A voter may see a cropped screenshot with the original label removed, watch a shortened repost that omits the notice, or encounter a copy shared in a private group without its original context. A clear label can improve accountability, but it cannot reverse every impression.

A local newsroom would face a related verification problem. Before embedding a disputed clip, reporters would need to locate the earliest available upload, contact the depicted campaign, examine any provenance data, and decide how much of the material could be shown without amplifying it. A small newsroom may have to perform that work while the clip is already spreading across group chats and social feeds.

An election office could face an even more immediate version of the problem. If a synthetic recording falsely tells voters that a polling place has closed, officials would need to correct the claim across their website, social accounts, local media, and telephone lines. A label attached to the original recording would offer little help if unlabeled copies were already circulating.

Speed compounds the problem. Campaigns can now generate many message variations and distribute them before opponents or newsrooms complete verification. By the time a correction arrives, the original content may have reached its intended audience.

These asymmetries make South Carolina AI campaign rules more than a question of limiting campaign creativity. Policymakers must decide who carries the burden when synthetic content creates confusion.

A rule could place that burden on speakers through mandatory labels. It could give depicted candidates a fast path to court. Platforms could be required to preserve provenance information, meaning records about where media originated and how it changed.

Each approach involves different enforcement costs and free-expression concerns. Doing nothing also allocates the cost. It places more responsibility on voters, journalists, election officials, and targeted campaigns after questionable material is already circulating.

South Carolina AI Campaign Rules Stalled at Disclosure

South Carolina’s principal proposal favored labeling deceptive media instead of imposing a complete ban on AI-assisted campaign speech.

House Bill 3517 was prefiled in December 2024, introduced in January 2025, and referred to the House Judiciary Committee. The South Carolina General Assembly’s official record for H. 3517 lists no legislative action after its January 14, 2025, referral during the 2025–2026 session. Because the measure did not advance into law, its provisions are described here as proposals rather than current legal requirements.

The bill focused on deceptive synthetic representations of candidates. It defined synthetic media as manipulated or generated images, audio, or video that created a realistic but false portrayal.

Its restrictions would have applied within 90 days of an election. Distribution would have been prohibited when a person knew, or should have known, that the content deceptively depicted a candidate on the ballot.

However, the proposal contained an important exception. Distribution could continue if the media included language stating that the image, video, or audio had been manipulated or generated by artificial intelligence.

The required notice was designed to remain visible throughout a video. An audio disclosure would have appeared at the beginning and end, with repeated notices at intervals of no more than two minutes in longer recordings.

H. 3517 also gave depicted candidates possible civil remedies. They could seek an injunction to stop publication and pursue general or special damages against a sponsor.

Criminal penalties were included as well. Under the bill text, a first violation could have brought a misdemeanor charge punishable by up to 90 days in jail, a fine of up to $500, or both. A second offense within five years could have been treated as a felony punishable by up to five years in prison, a fine of up to $1,000, or both.

Those provisions show that the proposal was narrower than a general South Carolina political AI law. It did not regulate campaign strategy, AI-assisted research, ordinary copywriting, or every generated image.

It targeted deceptive portrayals of candidates during a limited election window. Even then, a compliant disclosure could permit distribution.

The bill also specified exceptions for satire or parody and certain news, documentary, broadcasting, and periodical uses. Those exceptions matter because the proposal was written to address deceptive candidate portrayals without automatically treating reporting or obvious political comedy as unlawful.

That structure reflects a broader state trend. Disclosure rules try to preserve political expression while giving viewers information about how a message was made. They also avoid requiring the government to judge whether every political claim is true.

The bill’s lack of movement means South Carolina entered the late 2026 campaign period without that dedicated framework. The absence does not legalize fraud, voter suppression, defamation, or every form of impersonation.

Instead, it leaves enforcement dependent on laws written for other conduct. Those laws may address a harmful act without establishing consistent labels for AI-generated political media.

That distinction matters for voters. A deceptive robocall might violate telecommunications or voter-protection rules. A false advertisement might support a civil claim. Yet lawful synthetic content can still circulate without an AI notice.

South Carolina lawmakers must therefore decide whether the state needs a content-specific rule before another election cycle. They must also determine whether a 90-day window is sufficient when political material can persist online for years.

The proposed definition may require another look. It referenced generative adversarial networks, one method used to generate synthetic media, while also including “other digital technology.” Many newer systems rely on diffusion models or other technical approaches.

Legislation tied too closely to one method risks aging quickly. A more technology-neutral definition could focus on the resulting deception and material alteration while remaining precise enough to avoid covering routine editing.

Disclosure Offers a Middle Path, Not a Complete Defense

The central policy tradeoff pits transparent political speech against rules broad enough to chill satire, criticism, and legitimate production tools.

Disclosure has become the most common state response to deceptive AI election media. Rather than banning every synthetic message, states require a notice explaining that content was created or altered using AI.

By June 23, 2026, the National Conference of State Legislatures reported that 31 states had enacted laws regulating political deepfakes. Its state policy tracker identified two broad models: limited prohibitions and disclosure requirements.

Minnesota and Texas use time-limited prohibitions for certain deceptive media. Maryland applies a year-round prohibition, while 28 other states primarily rely on disclosures, according to the NCSL review.

This pattern reflects constitutional pressure. Political speech receives strong First Amendment protection, including some speech that is exaggerated, harsh, or false. In *United States v. Alvarez*, 567 U.S. 709 (2012), the U.S. Supreme Court rejected the proposition that false statements fall entirely outside First Amendment protection. The decision did not concern AI or election deepfakes, and it does not establish that every deepfake restriction is unconstitutional. It does illustrate why broadly written restrictions on false political expression can face constitutional scrutiny.

Satire creates one obvious boundary problem. A plainly comedic image may place a candidate in a fictional scene to make a political point. Treating it like a deceptive impersonation could sweep protected commentary into an election law.

Documentaries and news reporting create other complications. Journalists may need to show manipulated content while explaining why it is false. A workable rule needs exceptions without creating loopholes for campaign advertisers.

Disclosure is appealing because it does not require the state to approve political messages before publication. It asks speakers to identify synthetic material, much as existing campaign notices identify who paid for an advertisement.

However, labels vary in effectiveness. Small text, brief notices, or vague phrases can be technically present while remaining practically invisible. Audio and video need different standards because audiences consume them differently.

Reposting is another weakness. A notice can disappear when someone screenshots an image, trims a video, or uploads only part of a recording. Metadata can also be stripped by platforms.

For example, a voter might first encounter a synthetic video as a vertical social-media repost rather than as the campaign’s original advertisement. If the account crops the lower portion of the frame, a disclosure placed there may disappear. If the same voter later sees a still image taken from the clip in a private message, neither the original label nor the source account may remain visible.

Digital provenance systems can preserve information about a file’s origin and editing history. That data can help platforms and journalists evaluate media, but it does not guarantee that ordinary viewers will inspect it.

A resilient policy may therefore need both visible disclosure and machine-readable provenance. The visible notice serves voters, while embedded information supports verification and platform handling.

Even that combination cannot determine whether the message’s underlying political argument is honest. A real photograph paired with a false caption can mislead without containing generated pixels. A synthetic reenactment can communicate a documented event without pretending to be authentic footage.

Rules should consequently focus on material deception involving a real person’s appearance, actions, or speech. They should not promise to eliminate misinformation as a whole.

Campaigns also need internal controls. Generated drafts should remain linked to original sources, human approvals, and publication records. Reliable information capture can help teams preserve those references instead of relying on disconnected AI outputs.

In practice, that could mean retaining the original recording, the prompt or editing instructions, the staff member’s approval, and the final exported file. If a voice or scene is challenged later, the campaign would have a traceable record showing what was generated and who approved publication.

A useful test is whether someone outside the production team could reconstruct the message. For a synthetic video, the record should identify the source footage, generated elements, factual references, editor, approving official, disclosure text, and publication time. If the campaign cannot produce that chain, a newsroom or regulator has less evidence with which to distinguish an error from intentional impersonation.

That practice is not a substitute for law. It does reduce avoidable errors and gives campaigns evidence when questions arise about how a message was created.

For South Carolina, disclosure remains the clearest middle path considered in H. 3517. The difficult work lies in defining which content needs a label, how prominent that label must be, and who remains responsible after redistribution.

Federal Rules Cover Pieces of the Problem

Existing federal authority addresses specific conduct, but it does not provide a universal label for every AI-generated political advertisement.

The Federal Election Commission considered a request to create broader AI campaign-ad rules in 2024. It chose not to open the requested rulemaking.

Instead, the commission issued an interpretation explaining that existing federal restrictions on fraudulent misrepresentation can apply regardless of the technology involved. The rule covers particular situations involving false claims of speaking or acting for candidates or political parties.

The FEC’s September 2024 interpretation states that 52 U.S.C. § 30124 and 11 C.F.R. § 110.16 are technology-neutral and may apply to fraudulent misrepresentation carried out with AI-assisted media. The agency said it would apply the statute to specific technologies case by case. That authority is narrower than a general prohibition on deceptive campaign media and does not require every federal political advertisement containing an AI-generated element to carry an AI label.

The distinction is important because federal campaign law does impose sponsorship disclaimers on many communications. Those notices tell voters who paid for a message, but they do not necessarily explain whether its voice, image, or scene was generated.

The Federal Communications Commission took a separate action concerning calls. In February 2024, it ruled that AI-generated voices count as artificial or prerecorded voices under the Telephone Consumer Protection Act.

The FCC’s Declaratory Ruling, *Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts*, CG Docket No. 23-362, FCC 24-17, established that calls using those voices fall within the TCPA’s existing restrictions. It followed a New Hampshire incident involving an AI-generated voice resembling then-President Joe Biden and gave state attorneys general another legal path for pursuing certain unlawful robocalls.

The ruling did not prohibit every AI-generated political call in all circumstances. Robocall legality also depends on consent, purpose, and applicable statutory exceptions. Still, it demonstrated that existing statutes can reach new technical methods.

For a recipient, that fragmented protection produces different practical signals. A call may be regulated because it uses an artificial voice, while an online video containing the same cloned voice may fall under a different set of federal, state, platform, and civil rules. The underlying technology can be similar even though the enforcement path changes with the delivery channel.

A campaign compliance team would therefore need to ask at least four separate questions before distribution: who paid for the communication, whether it falsely claims to speak for another candidate or party, whether the delivery method triggers telecommunications rules, and whether a state synthetic-media law requires a label. Passing one test does not establish compliance with the others.

This fragmented approach leaves gaps between agencies and jurisdictions. The FEC regulates federal campaign finance, not state and local elections as a whole. The FCC governs communications systems, not every online video or image.

State election laws therefore remain central. They can address communications involving state and local candidates, specify disclosure formats, and provide local procedures for rapid relief.

Yet a state-by-state system produces inconsistent expectations. A campaign advertisement may be legal without a label in one jurisdiction and require a specific notice in another. Online distribution rarely respects state boundaries.

Platforms must decide whether to apply the strictest state rule nationwide, target notices geographically, or build different handling procedures for each jurisdiction. Small campaigns face the same compliance problem with fewer lawyers.

Federal baseline legislation could reduce that inconsistency, but it would still need to preserve room for legitimate political speech. It would also need to distinguish misleading impersonation from routine editing and harmless generative assistance.

Until that happens, South Carolina sits inside a patchwork. Existing law can punish some harmful conduct, but the state has not adopted a comprehensive disclosure standard for synthetic election media.

That uncertainty affects more than enforcement after an incident. It shapes campaign planning before publication. Teams must decide what to disclose voluntarily, even when state law does not clearly require it.

A practical campaign should assume that unlabeled synthetic portrayals create reputational risk. Legal permission does not guarantee voter acceptance, platform distribution, or favorable press coverage.

Voluntary labels can establish a norm before lawmakers act. They can also reveal where labels work poorly, giving policymakers evidence for future standards.

The Next Test Is Whether Rules Can Move as Fast as Campaigns

The debate will turn on three observable developments: renewed state legislation, campaign disclosure practices, and enforcement actions involving deceptive content.

First, lawmakers can revisit South Carolina AI campaign rules with lessons from other states. A future proposal could define deceptive synthetic media by its effect rather than relying primarily on a single generation method.

A renewed bill would also need a clear scope. Legislators must decide whether the rule applies only to candidates, or also covers election officials, ballot questions, and fabricated events intended to suppress voting.

Timing deserves similar scrutiny. A 90-day window concentrates enforcement near an election, when harm may be greatest. It also leaves earlier synthetic content outside the dedicated rule, even if that material remains online during voting.

If lawmakers reintroduce a disclosure proposal, its progress will show whether the General Assembly is prepared to reconsider the issue. Committee hearings would also make enforcement concerns, technical definitions, and constitutional objections part of the public record.

Second, campaigns can adopt voluntary standards before a mandate exists. Those standards should separate internal assistance from public impersonation and require a prominent label when synthetic media realistically depicts a person.

The relevant measure is not how many campaigns say they use AI responsibly. It is whether their published advertisements contain consistent, understandable notices that survive ordinary sharing.

Campaigns should also preserve the real source material behind generated scenes or quotations. Human review should verify claims before publication, particularly when content represents an opponent’s words or actions.

Third, enforcement will reveal how existing laws operate in specific cases. A late synthetic attack could implicate defamation rules, voter-suppression statutes, campaign-finance requirements, or federal communications law, depending on its content, speaker, distribution method, and effect.

Whether existing procedures provide timely relief will become clearer only through official enforcement records and court decisions. If disputes repeatedly fall outside current statutes or move too slowly to address election-period harm, lawmakers will have more concrete evidence to consider when debating a dedicated rule.

Actual disputes will also clarify the difference between deception and obvious parody. Courts, election officials, and platforms may reach different conclusions, which would increase pressure for a clearer statutory standard.

Voters should watch the labels as carefully as the content. A disclosure must remain readable, audible, and attached to the media. A notice hidden in a caption or removed during reposting offers little protection.

A reader evaluating a questionable clip can take several concrete steps: look for the earliest upload rather than a repost, check whether the candidate or election office has addressed it, compare the voice or footage with an authenticated recording, and consult reporting that explains how the media was verified. None of these steps proves authenticity alone, but together they provide more evidence than visual realism does.

Newsrooms face a related responsibility. Repeating a fabricated clip while debunking it can extend its reach. Reporting should foreground verification, show only what is necessary, and avoid presenting synthetic media without context.

Platforms can contribute by retaining provenance data, providing visible warnings, and creating faster channels for election officials and depicted candidates. Those systems need transparent appeals so that political criticism is not removed merely because it uses edited media.

The larger challenge is cultural as well as legal. As synthetic content becomes common, people may distrust genuine recordings. Researchers and legal scholars often describe this as the “liar’s dividend”: the possibility that authentic evidence can be dismissed merely by alleging that it was generated or manipulated.

Disclosure rules address only one side of that problem. Provenance records, trusted reporting, campaign accountability, and public skepticism must work together.

South Carolina does not need to prohibit the productive use of generative AI to protect voters. It can distinguish ordinary assistance from deceptive impersonation and give people meaningful context before synthetic media shapes their understanding.

The state’s next legislative move will indicate whether disclosure becomes part of that response. Until then, every campaign has a choice: use AI only because it is inexpensive, or use it with records and labels that voters can evaluate.

That choice will shape political trust before any statute does. Voters, campaigns, and lawmakers should ask the same factual question whenever synthetic media appears: does this message clearly identify its origin and alterations, or does it depend on people mistaking generated content for reality?