Protecting Consumers from Deceptive AI Act

This bill requires the National Institute of Standards and Technology (NIST) to create task forces to develop technical standards and guidelines for identifying content made by generative artificial intelligence (AI), and mandates that audio or visual content created or substantially modified by generative AI include a disclosure of its AI origin, enforced by the Federal Trade Commission (FTC).

Sec. 3, Guidelines to facilitate distinguishing content generated by generative AI: Within 90 days, the NIST Director must establish task forces to support technical standards for content provenance metadata, watermarking, and digital fingerprinting for audio or visual content, and for identifying and labeling text-based content from generative AI. The task forces include representatives from federal agencies, AI developers, standards organizations, social media platforms, search engines, academics, privacy advocates, and others. They must report to the NIST Director within 270 days and to Congress annually for five years, and consider privacy guidance.

Sec. 3(b), Informing consumers of content generated by AI: Providers of generative AI applications must ensure audio or visual content includes a machine-readable disclosure of AI origin, prevent tampering, collaborate with online platforms, and allow users to add metadata about the AI origin. Providers of covered online platforms (with at least $50 million annual revenue or 25 million monthly active users) must display such disclosures and not remove them. The FTC must issue regulations within two years, and violations are treated as unfair or deceptive acts. Compliance can be achieved through FTC-approved self-regulatory guidelines, with appeals to federal court. The FTC must consider privacy and interoperability.

Sec. 3(c), Definitions: Defines key terms including 'audio or visual content' as digital images, video, or audio; 'covered online platform' as websites or apps in the U.S. meeting revenue or user thresholds; 'generative artificial intelligence' as models using deep learning to generate new data; and 'watermarking' as embedding tamper-resistant information into digital content.