Intentional AI interference with voting rights would carry criminal penalties

Developers and users covered by the bill would be barred from intentionally using AI or similar tools to deny or cheat people out of voting rights. The rule would cover federal, state, and local elections, including deception or intimidation involving voting, candidate support, and election work. The conduct must intentionally deny or cheat someone out of voting rights, or attempt to do so. The exclusion for individual users acting for noncommercial purposes does not apply to developers. The prohibition and penalties would take effect 60 days after the bill becomes law.

Publishers covered by the bill would have to include a written AI notice or a circled AI symbol. The rule would cover simulated or altered candidate images, videos, or voices and voting or registration instructions in federal, state, and local elections. The publisher must intentionally publish content generated wholly or partly with a covered tool. The Federal Trade Commission could exempt content that a reasonable person would clearly recognize as artificially generated. The waiting period would begin when the Federal Trade Commission issues the required labeling rules.

Developers would have to provide tools that let publishers add the required AI labels. This duty would cover developers offering tools that generate or publish simulated or altered candidate images, videos, or voices or voting and registration instructions. Developers would have to make the labels permanent or hard to remove to the extent technically possible. The waiting period would begin when the Federal Trade Commission issues the required labeling rules.

Developers and users covered by the bill would have to meet safety and performance rules when offering, licensing, or using covered software. These duties would apply beyond election uses and would cover developers using their own tools. They would have to take reasonable steps to reduce identified harms and consider not offering or using the software. They must reasonably conclude that major harm and deception are unlikely, benefits outweigh costs, and people can reasonably avoid being affected. They would need reasonable steps to meet expected and advertised performance and match advertised uses.