Can the government force journalists to disclose their use of generative artificial intelligence (GenAI) tools to help compose or create news stories? Would such a transparency mandate—one aimed at promoting trust in journalism (GenAI models sometimes hallucinate) and safeguarding jobs (“human news work”)—raise First Amendment concerns? These aren’t idle questions.
Lawmakers in the Empire State this month overwhelmingly passed companion bills proposing the “New York Fundamental Artificial Intelligence Requirements in News Act” (FAIR News Act). The legislation requires “any news media content” that is “substantially composed, authored, or otherwise created through the use of generative artificial intelligence” to “conspicuously” disclose such usage when it is published, broadcast, or simply made accessible in New York. Covering a range of outlets—newspapers, magazines, websites, and broadcast and cable stations, among others—the Act allows New York’s Attorney General to seek injunctive relief and civil fines.
The Act, which now is before Gov. Kathy Hochul, mingles policy interests with constitutional concerns that will arise again, regardless of whether Hochul signs it, because its language is expected to be “introduced in other statehouses.”
Transparency and Trust. Patricia Fahy, the Act’s Senate sponsor, says the measure is partly intended “to protect the public’s trust in reporting at a time when trust in media and reporting is at an all-time low.” Nily Rozic, who sponsors it in the Assembly, adds that “as artificial intelligence continues to reshape how news is produced and distributed, we have a responsibility to ensure transparency.”
Transparency and trust are important journalistic concerns. One of four overarching principles articulated by the Society of Professional Journalists in its Code of Ethics is that journalists should “be accountable and transparent.”
Lack of trust in journalism is a major problem, although its magnitude seemingly depends on one’s political affiliation. The Pew Research Center reported in February that “a majority of Americans (57%) express low confidence in journalists to act in the best interests of the public.” Unpacking the data, however, reveals a sharp political divide: “Democrats and Democratic-leaning independents (61%) are more than twice as likely as Republicans and GOP leaners (25%) to say they have confidence in journalists to act in the best interests of the public.”
Whether disclosing substantial GenAI usage in composing, authoring, or creating news stories would ameliorate this situation is doubtful. One suspects that some people’s distrust of journalists long preceded the rise of GenAI tools and that such distrust is too deeply ingrained to be cured by a GenAI warning.
First Amendment. R Street’s Spence Purnell points out that the legislation’s mandatory disclosure requirements push the Act “into constitutionally treacherous territory. The Supreme Court has long held that compelled speech is generally as offensive to the First Amendment as suppressed speech.” Indeed, the Act would be judicially reviewed under the hard-to-pass strict scrutiny standard.
The Supreme Court recently observed that the government generally “may not compel a person to speak its own preferred messages.” The constitutional right not to speak safeguards newspapers as well as people, as the Court recognized in Miami Herald Publishing Co. v. Tornillo. The justices in Tornillo emphasized that “the choice of material to go into a newspaper” involves “the exercise of editorial control and judgment” protected by the First Amendment.
Diane Kennedy, president of the New York News Publishers Association, notes that the FAIR News Act troublingly “enables the state’s attorney general to investigate how news stories are gathered and reported.” The New York State Broadcasters Association similarly “warns the legislation injects the government into every newsroom in the state by giving the Attorney General the authority to examine every newsroom to see if a news story needed a disclosure.”
Due Process. Laws affecting speech that don’t clearly define key terms face due process challenges under the void for vagueness doctrine because they fail to provide adequate notice about when they apply. An obvious problem with the Act is that it doesn’t explicate what “substantially composed, authored, or otherwise created through the use of” GenAI means. (emphasis added). In short, how much GenAI input is too much isn’t clear. This vagueness, in turn, leaves New York’s Attorney General with vast leeway and discretion to decide what’s substantial when enforcing the Act.
Copyright Carveout. The Act exempts from coverage “content [that] is eligible for copyright registration.” Works wholly generated by AI tools currently don’t receive copyright protection because they lack human authorship. Conversely, the US Copyright Office holds that “neither the use of AI as an assistive tool nor the incorporation of AI-generated content into a larger copyrightable work affects the availability of copyright protection for the work as a whole.” This exemption could safeguard from the Act’s reach many news stories composed with some GenAI assistance, but it won’t prevent the Attorney General from intrusively investigating how those stories are composed.