Transparency in the Age of AI: The Role of Mandatory Disclosures
Policymakers face a tough challenge when it comes to regulating the fast-moving AI industry. However, disclosure requirements are one tool in their regulatory toolbelt that policymakers have embraced in the early stages of AI development. By mandating that you must disclose when you’ve used AI to generate media or an AI tool, policymakers thread the needle of providing useful information to consumers while not placing a substantial roadblock in front of the nascent industry’s development. Recent reports of low-quality, AI-generated books and articles flooding the market highlight the need for consumers to know which products were created by people versus AI. Disclosure requirements are a logical first step as we all become accustomed to AI in our everyday lives.
The most straightforward case for mandatory disclosures on AI-generated content is in political campaigns, especially if a political candidate’s image or voice is manipulated by AI and spread through social media. Voters already see disclosures in campaign ads under existing rules requiring a disclosure identifying the group responsible for political advertisements. States have been very active in requiring similar disclosures for AI-generated political advertising. Last year, Washington enacted a bill (WA SB 5152) to require a disclosure when any manipulated audio or visual media is used in an electioneering communication and Michigan enacted a law (MI HB 5141) requiring a disclaimer (“This message was generated in whole or substantially by artificial intelligence”) for pre-recorded phone messages and political advertisements that were created with AI within 60 days of an election. Already this session, we’ve seen similar bills introduced in Alaska (AK SB 177), Florida (FL HB 919, HB 1459, SB 850, & SB 1680), Nebraska (NE LB 1203), New Hampshire (NH HB 1500), Oklahoma (OK SB 1655), South Dakota (SD SB 96), and Wisconsin (WI AB 664 & SB 644)
While we haven’t seen mandatory disclosure laws enacted outside of the political context yet, this concept can easily expand to other areas. Lawmakers in California (CA AB 1824), Massachusetts (MA HD 4788), and Pennsylvania (PA HB 1598) have introduced legislation to require the disclosure of content generated through AI in general. Other bills require an additional qualification for the AI-generated content to require disclosure. For example, a bill in Virginia (VA SB 164) requires disclosure as long as the AI-generated content “depicts an actual person,” bills in Illinois (IL HB 3285) and Wyoming add the requirement that the AI-generated content must be misleading, and a New Jersey bill (NJ AB 1604) requires disclosure only when the AI-generated content is “intentionally deceptive.”
Additional disclosure bills are limited to a specific medium. A New York bill (NY AB 8098 & SB 7922) would require disclosure only for books created wholly or partially with the use of generative AI and another bill (NY AB 8158 & SB 7847) would require disclosure for AI content in newspapers, magazines, and “other publications printed or electronically published” in New York. Finally, bills in Washington (WA SB 6073) and New York (NY AB 8110) would require disclosure when AI generation is used in legal proceedings.
Beyond AI-generated media, disclosures are also a key aspect of proposed regulations on AI decision tools. In this context, a company using AI tools for decision making affecting a consumer would be required not only to disclose the use of AI in that decision but also proactively notify the consumer of that use. California’s draft regulations proposed such a mechanism as well as major legislation (CA AB 331) introduced by Assemblymember Rebecca Bauer-Kahan (D), which would require notification of any person subject to a “consequential” decision by an AI tool.
The difficulty will be where to draw the line when defining when a disclosure requirement is triggered. For example, Michigan’s AI in elections law requires a disclaimer when media “is generated in whole or substantially” by AI. But that “substantially” qualifier will require some interpretation. If a writer uses AI to brainstorm ideas for a newsletter or asks AI to generate the article’s title, would that trigger a disclosure? What if an artist incorporates an AI-generated image into their own artwork, which is further modified by the artist? If a movie studio uses AI tools to edit a film shot without the help of AI, would that require disclosure?
Regardless of the answers to the above questions, we could see diminishing benefits of disclosure in the long run if AI is incorporated into more of our daily tasks. We could eventually become numb to a “generated with AI” disclosure as they become more ubiquitous in the same way that most people ignore the “known to cause cancer” warnings on everyday products mandated by California’s Prop 65.
But for now, state lawmakers have embraced disclosure as a way to promote transparency as AI technology is infusing itself into our everyday lives. Expect lawmakers to utilize these types of disclosure and notification requirements as they expand their regulatory scope over AI technologies.
Recent Policy Developments
Florida: Lawmakers have already advanced a number of deepfake bills early in the session. The House Civil Justice Subcommittee amended and approved legislation (FL HB 757) that would provide liability for defamation if AI is used to deceive the public into portraying someone in an offensive light that a reasonable viewer would believe is real. Additional legislation requiring disclaimers for deepfakes in political ads (FL HB 919 and SB 850) were passed out of committee in each chamber, although concerns were raised about font size in the disclaimer.
Hawaii: Lawmakers introduced the first artificial intelligence bill in the Aloha State (HI HB 1607), proposing legislation on algorithmic decision making that would prohibit discrimination, require notice to individuals subject to the technology, and require regular audits with results reported annually to the Attorney General. The proposed law would only apply to data brokers, service providers, or companies that possess personal information of at least 25,000 residents, or have more than $15 million in annual gross receipts.
New York: The Department of Financial Services has posted for comment a “circular letter” offering artificial intelligence guidance for insurers. The letter calls for the establishment of a proper governance and risk management framework, offers measures to reduce the risk of unlawful discrimination, and calls for greater transparency.
North Carolina: Attorney General Stein (D) was part of a group of 25 attorneys general calling on the federal government to restrict telemarketers from using artificial intelligence to make robocalls. They argue AI cloning of human voices constitutes a form of "artificial voice" under the federal Telephone Consumer Protection Act, violating provisions that prohibit such calls without the consumer's prior express written consent.
Pennsylvania: Gov. Shapiro (D) announced that his Office of Administration has signed an enterprise agreement with OpenAI for state employees can pilot the use of ChatGTP to assist with regular work tasks. “The pilot is the commonwealth’s first-ever use of generative AI tools for state government employees, and its findings will help guide the wider integration of this technology into state government operations to enhance productivity and empower the Commonwealth’s workforce.”