In my February Radio World article “Thoughts on AI and the Law,” I suggested that every station using artificial intelligence should adopt a policy addressing AI use disclosure, consent and compensation, and should obtain certifications from programmers and advertisers concerning their use of AI.
California’s recent legislation — and the rapid spread of similar laws around the country — makes that recommendation more urgent.
That the industry is moving forward with AI is inescapable. However, broadcasters must be careful and open about its use and cannot simply treat it as just another cost-saving production tool. It must be recognized as an issue that can impact your reputation for honesty, and compliance with political broadcast regulations.
California has addressed these issues perhaps more comprehensively than any other state and serves as a good model for broadcasters nationwide, although many other states appear to be following the same path.
California’s framework
In 2026, the state created a multi-level legal structure that affects many aspects of AI use by broadcasters.
Its regulations address a variety of issues, including synthetic-content origin and changes in its use, known as “provenance,” training, frontier-model safety, mitigating advanced AI risks, privacy and employment discrimination.
Effective in August, the California AI Transparency Act, or CAITA, requires producers of publicly accessible generative-AI with more than 1 million monthly users or visitors to provide a free public detection tool and to embed disclosures in AI-generated images and video and audio content that permanently includes the content’s origin and history.
The law reaches beyond AI platforms. Beginning in 2028, manufacturers of equipment used by broadcasters, such as cameras, mobile phones and microphones, among other things, must automatically embed origin and creation information (called “latent disclosures”) into the recorded content.
In addition to CAITA, California has enacted several statutes and regulations that touch on AI and are now in effect. These address a myriad of issues, including training requirements, safety risk assessments, critical-incident reporting, whistleblower protection and AI-assisted employment decisions.
While most individual stations will not be regulated as frontier-model developers or covered GenAI providers, they need to be aware that these laws affect the tools they buy and that the provenance information (where an object, data or document comes from and how it has changed over time) attached to the AI tools they may use.
Many observers believe that California’s AI laws may influence laws in other states and even, potentially, federal or industry standards.
California political
With campaign season upon us, it is important to note that California has already adopted disclosure requirements for political ads with AI.
The rule applies to California political committees that create, originally publish or originally distribute qualifying political advertisements. Any advertisements that contain an image, audio, video or other media that was entirely created by AI, and would falsely appear authentic to a reasonable person or was materially altered by AI in a way that would give a reasonable person a fundamentally different understanding of the material, must include, in a clear and conspicuous manner, the following statement: “Ad generated or substantially altered using artificial intelligence.”
This disclosure applies to radio and television broadcast ads that contain AI.
For radio, a political spot containing a cloned candidate voice, a fabricated exchange or other qualifying synthetic audio must carry a spoken disclosure. A separate requirement applies to television and online video content governing how the required AI disclosure must be placed and appear.
These disclosure requirements are imposed primarily on the political committee rather than a station that runs it, but stations should be knowledgeable and be certain to inquire of the producer or distributor whether AI was used, so that the broadcaster does not find itself inadvertently airing unlawful content or damaging its credibility.
Idea gone national
While California is an important example to study, it is far from alone in regulating AI use in political communications.
Thirty-one states have enacted laws regulating the use of deepfakes in political and election communications, according to tracking from the National Conference of State Legislatures.
Most state laws include disclosure requirements similar to California. Minnesota and Texas prohibit the publication of political deepfakes a certain number of days prior to an election. Maryland has no time limit and instead prohibits deceptive deepfakes related to the election year-round.
The other 28 states require disclosures on the media, similar to the existing requirement to disclose who paid for a political ad, the AI disclosure requires the ad to state if it contains a deepfake.
Colorado and Utah require additional disclosures in the metadata — descriptive information about a file’s creator, when the file was created and when the file was edited. Some provide civil remedies, while others carry criminal penalties.
This developing regulatory patchwork presents a practical problem for station groups and political buyers. A spot prepared for national or regional distribution may comply with one state’s laws, but not another. The required words, placement and duration may differ. So may the definitions of “AI-generated,” “synthetic media,” “deepfake” and “material alteration.”
Broadcasters therefore should no longer accept a general representation that a political advertisement is “legally compliant.” The representation should address the states and platforms in which the spot will run and should specifically identify whether AI was used to create or materially alter any voice, image or video.
The Section 315 problem
Political broadcasting law makes compliance with new AI regulations even more complicated for broadcasters.
They retain discretion and can censor noncandidate issue, independent-expenditure and ballot-measure advertisements and can demand corrections. However, a legally qualified candidate’s “use” presents a different problem since the Communications Act prohibits a station from censoring a candidate use.
The Communications Act states clearly at § 312(a)(7) that all legally qualified and bona fide write-in candidates for federal elective office are to have “reasonable access” to broadcast facilities. Those federal requirements do not fit neatly with state laws requiring specified words or visual elements in AI-generated political advertising as the broadcaster is prohibited from censoring the ad and must run them for a candidate for federal elective office provided they contain a “use” by the candidate.
A “use” is defined as any positive use of a candidate’s voice or picture in a context not otherwise exempt under Section 315 of the Communications Act. Thus, any program or commercial that identifies the candidate by voice or picture in a positive way and is not a bona fide newscast, bona fide news interview, bona fide news documentary or on-the-spot coverage of a bona fide news event, cannot be censored by the broadcaster. As a result, the broadcaster is provided immunity from liability for the content of the ad.
However, it cannot be assumed that Section 315 automatically disposes of every question involving state law or station liability. Therefore, to the extent possible, the safest course is prevention. Ask the whether AI was used when the order is placed, explain the applicable disclosure requirements to the buyer and ask that the candidate or committee deliver a compliant spot.
The FCC considered a federal solution in 2024. In MB Docket No. 24-211, the commission released a Notice of Proposed Rulemaking that asked whether the FCC should increase transparency in political ads by having political advertisers indicate if AI is being used in the advertisement and make on-air disclosures of AI use. It does not propose the prohibition of such content, only disclosing the use of AI within political ads. It further proposed to require broadcasters and certain other video and audio distributors to inquire whether political advertising contained AI-generated content, make an on-air disclosures and place a corresponding notice in their online political file.
The FCC explained it was attempting to address consumers’ concerns that AI-generated content was leading to a rising tide of disinformation, while also bringing uniformity to the patchwork of state laws and greater transparency in our elections.
The docket is still active with a list of filed comments 49 pages long, but the FCC has not yet acted. So, for now, broadcasters must continue to comply with state-by-state requirements as well as the established FCC sponsorship identification and political broadcasting rules.
A broader risk
While the focus is on political advertising in election years, there is a longer-term regulatory risk.
AI-generated voices can duplicate the authentic voices of real people, including a station personality, public official, celebrity or ordinary caller. It can “hallucinate” an apparently authentic interview, eyewitness account, endorsement or news clips. This can lead to increased liability for stations that are using AI to help generate content.
Existing law already reaches much of this conduct. Defamation, false light, the right of publicity, copyright, fraud, deceptive-advertising law and contract rights do not disappear merely because an AI model produced the content. A disclosure that the production of a segment used AI will not cure a defamatory statement, an unauthorized voice clone or a fabricated emergency announcement.
Stations should mandate human editorial review for news, elections, weather emergencies, health information, public safety and any content purporting to quote from or depict a real person. An “AI-assisted” label cannot become a substitute for human verification of factual statements.
The same caution towards AI use applies to entertainment and promotional programming. A synthetic host may be efficient, but the station must determine whether it has the right to reproduce the voice or persona on which the host is based.
Talent agreements should expressly address training, cloning, reuse, derivative voices, post-employment use and compensation. The SAG-AFTRA principles of consent, disclosure and appropriate compensation remain a sound guide, even when a particular employee or contractor is not covered by a collective bargaining agreement.
Stations should also examine syndicated programs, barter programming, paid interviews, commercials and network feeds for AI use. A station often will not know whether outside material was produced using AI unless it asks. Programmer and advertiser agreements should require disclosure of material AI use and representations that all necessary rights and consents were obtained.
Digital distribution changes the analysis
Broadcasters should resist the temptation to treat their website, app, podcast, streaming channel, connected-TV product and social-media accounts merely as extensions of the over-the-air station. They may be governed differently under relevant law.
As noted above, since Aug. 2, 2026, CAITA has required certain very large providers of publicly accessible generative-AI systems to provide detection tools and content-provenance disclosures. Most local stations that use third-party AI services will not be covered by these requirements, but the analysis could change if a media company develops, substantially modifies or publicly offers its own generative-AI service.
Even where CAITA does not regulate the station directly, it offers broadcasters a new resource to verify compliance. Newsrooms can use provider detection tools and examine provenance information when evaluating suspicious audio, photographs or video. Stations also should preserve — rather than strip — standards-compliant metadata and digital signatures as material moves through editing, traffic, web-publishing and social-media systems.
The 2027 CAITA very large-platform requirements will make that preservation practice even more important for a station group whose digital operation could satisfy the statutory platform thresholds.
California’s Training Data Transparency Act only applies to developers that make generative-AI systems publicly available. A broadcaster that merely uses a commercial AI service ordinarily is a customer, not the developer. A broadcaster that trains and releases a public model using its news archive, announcer voices, photographs or programming may be in a very different position. It also must determine whether it has the rights necessary to use that material for training.
Digital operations raise privacy questions as well. Station apps and websites may collect personal identifiers, location information, listening histories, contest data and advertising profiles. California’s privacy rules now address certain automated decision-making and risk assessments. AI used for targeted advertising, audience profiling, hiring or other consequential decisions can therefore create obligations that ordinary over-the-air broadcasting does not.
A policy for 2026
Given the expanding state-law patchwork, it is essential that each station have a written AI policy in place. At a minimum, each station should:
- Require political buyers and outside programmers to identify AI-generated or materially AI-altered content in writing.
- Require each buyer to certify compliance with every applicable state disclosure law.
- Route candidate uses that do not comply with AI disclosure requirements to counsel rather than editing material at the station to attempt to comply.
- Require human review of news, emergency, election, health and public-safety content.
- Prohibit unauthorized cloning of employee, candidate, celebrity or other persons voices or likenesses.
- Preserve content credentials, watermarks and provenance metadata rather than stripping them.
- Limit the entry of confidential station, employee, advertiser and source information into public AI systems.
- Address ownership, consent, compensation and post-employment use in talent and vendor agreements.
- Include separate policy provisions that apply to broadcast, streaming, websites, apps, podcasts and social media.
California’s most important message is not the wording of one disclaimer. It is that broadcasters are entering an era in which they will be expected to know — or at least to ask — whether the voices and information they distribute are real.
AI can help broadcasters remain efficient, competitive and creative. But broadcast’s most valuable asset has never been its ability to generate more content at lower cost. It is the audience’s belief that the station knows and serves its community and can be trusted. Any AI policy should begin there.
This column is provided for general information purposes only and should not be relied upon as legal advice pertaining to any specific factual situation. Legal decisions should be made only after proper consultation with a legal professional of your choosing.
