Large AI services would mark synthetic content and reveal its source

Covered providers would let users add a clear AI label and would add hidden details to content their AI systems create or change. These rules cover consumer-facing services, websites, apps, or interfaces that let users create or substantially change content with generative AI, have over two million active monthly users in the United States, and are publicly accessible in New Mexico. The clear label must be permanent or very hard to remove when technically feasible. Hidden details must be detectable by a provenance tool and follow accepted standards. Providers that license their systems must require continued compliance and revoke a license within 96 hours after finding a major failure.

Device makers would add hidden capture details by default and let users turn them off. Large platforms would show source details and provide a way to request removal of deceptive synthetic content. Platform rules cover public-facing social, file-sharing, mass-messaging, and search services with over two million active monthly U.S. users that share synthetic content. Broadband and phone services are excluded. The device requirement applies to devices first produced for sale in New Mexico on or after January 1, 2028; following accepted standards is required when technically feasible. Platforms must respond to removal requests within 72 hours. Hosting alone does not count as knowingly sharing deceptive content if a platform in good faith preserves source details, responds promptly, and does not materially spread the content after a request. The device requirements apply to devices first produced for sale in New Mexico on or after January 1, 2028.

The attorney general could investigate violations and seek civil penalties. A person alleging a violation could sue in district court. Deceptive synthetic content must realistically depict conduct the identifiable person did not engage in, be shared without that person's consent, and be reasonably believable as authentic. Sharing deceptive synthetic content creates civil liability only when the person knows or recklessly disregards that sharing is reasonably likely to cause harassment, entrapment, defamation, extortion, or other financial or reputational harm. The liability rule excludes specified lawful activities and certain satire, parody, criticism, teaching, research, education, or news reporting when a reasonable person would not believe the content accurately shows the depicted person's speech or conduct. A winning plaintiff may receive attorney fees and the greater of $1,000 per view or interaction or actual damages. The attorney general may seek $15,000 per violation.

A separate court or jury finding that generative AI helped commit a noncapital felony would add one year to the prison sentence. This applies when a generative AI system was used to assist in committing a noncapital felony. The extra year would be served first and could not be suspended or deferred.