Election AI Deception Prevention

Summary#

H.R. 10841, the Securing Elections From AI Deception Act, would create a new federal law; it does not amend a named existing Act. Its stated goal is to stop AI from being used to deprive or defraud people of the right to vote. The bill is at an early stage and is not law; whether it will pass is uncertain.

• Developers and deployers of covered algorithms would be barred from intentionally using them to deceive or intimidate people in ways that interfere with voting, voter registration, election support, or election workers’ duties. A violation could bring a fine, up to one year in prison, or both.
• Deployers who intentionally publish certain AI-generated election-related content would have to add an AI disclaimer. This covers simulated or altered candidate images, video, or voices, and content about voting procedures or requirements.
• Developers offering tools to make or publish that content would have to provide a way to add the disclaimer and, when technically feasible, make it permanent or hard to remove.
• The Federal Trade Commission (FTC) would enforce the disclaimer rules and a new duty of care for covered algorithms. That duty would apply broadly, not just to election-related uses.
• The FTC would have to issue disclaimer rules within two years of enactment. The disclaimer requirement would start 60 days after those rules are issued; the voting-deception prohibition would start 60 days after enactment.

What it means for you#

• Voters and election workers: The bill would prohibit intentional AI-enabled deception, threats, intimidation, fraud, or coercion intended to interfere with voting, registration, candidate support, or election work. It would not change how people register or cast ballots.
• People and organizations publishing election-related content: If a deployer intentionally publishes content made in whole or in part with a covered algorithm, a disclaimer would be required for the listed candidate simulations or voting-procedure information. The bill allows either a written statement or an “AI” symbol. FTC rules would set details such as placement and accessibility.
• AI developers and deployers: They could not offer, license, or use a covered algorithm in a way that is not “safe and effective.” To be safe, they must take reasonable steps to address harms identified by a pre-deployment evaluation or impact assessment; ensure intended use is unlikely to violate the Act; and evaluate whether not offering, licensing, or using the algorithm—or removing it from use—is preferable. They must reasonably conclude that use is unlikely to cause substantial harm or deception, that benefits outweigh costs to affected people, and that people can reasonably avoid being affected. To be effective, they must take reasonable steps on performance, consistency with expected and advertised performance and purpose, and whether data is relevant and appropriate to the context and purpose.
• People using tools in a non-commercial context: They are excluded from the bill’s definition of a “deployer.” The bill does not clearly exclude an individual who meets its separate definition of a developer.
• The FTC: It would have to make rules, educate the public about the disclaimer for five years, and report to Congress on the disclaimer’s adoption and effectiveness. The bill does not name an enforcement agency for the criminal voting-deception prohibition.

Money#

No publicly available information.

• The bill gives no cost estimate or specific appropriation.
• The FTC would have to issue rules and conduct public education and reporting. Developers and deployers would have compliance duties, but the bill gives no estimates of those costs or required staffing.

What is unclear#

• “Covered algorithm” includes a wide range of computational processes, but the bill does not clarify how its scope would be applied in particular cases.
• The bill does not set out how pre-deployment evaluations or impact assessments must be done, or require a specific format, reviewer, or reporting process.
• The bill does not set a separate start date for the duty of care in section 5.
• The bill does not identify who would investigate or prosecute violations of the criminal prohibition.
• The bill does not say how the new duties would interact with other rules or programs that may apply to the same conduct.

Case for#

• The bill targets deliberate deception about voting and deliberate efforts to deter voters or interfere with election workers.
• Disclaimers could help people recognize when certain candidate images, voices, or voting information were made with AI.
• A duty to assess harms and performance could push developers and deployers to address risks before and during use, rather than only after harm occurs.
• FTC rules could set common requirements for disclaimer placement and accessibility across different media.

Case against#

• The duty of care covers a broad range of algorithms and is not limited to election uses. Unclear terms such as “safe,” “effective,” and “reasonable” could make it hard to know what compliance requires.
• The bill sets no detailed method for assessments or clear process for showing that an algorithm meets the duty of care.
• The disclaimer requirement may be difficult to apply consistently across media. Developers need only make disclaimers hard to remove when technically feasible, and the bill allows the FTC to create exceptions.
• The bill assigns new rulemaking, education, and reporting work to the FTC without stating the resources or costs.