S. 4915: AI Labeling Act of 2026

This bill would require certain artificial intelligence systems and major online platforms to label AI-generated images, videos, and audio in a clear way.

What has to be labeled

The bill focuses on “covered AI-generated content”, meaning digital content such as images, videos, or audio that is created or substantially changed by generative AI in a way that would matter to a reasonable viewer, and that would not obviously appear to be AI-made.

Providers of generative AI systems that produce this type of content would have to:

• put a clear and conspicuous label on the content saying it includes AI-generated content;
• make the label accessible to people with disabilities, when technically and economically feasible;
• embed the label in the content itself;
• include a machine-readable disclosure showing, at minimum, the content’s AI origin, the system and version used, and when it was created or modified;
• make the content detectable by common detection tools, or provide a tool if no widely available tool exists; and
• work with covered online platforms to help those platforms comply with the bill.

Rules for online platforms

Large public-facing websites and apps that mainly host user-to-user content, search, or content aggregation would have to:

• clearly identify covered AI-generated content displayed on the platform;
• not remove or alter the disclosure when content is shared or transferred to another platform;
• let users choose to share provenance information, including the content’s AI history, with others;
• make a good-faith effort to reduce false claims that real content is AI-generated; and
• make the disclosures accessible to people with disabilities when feasible.

Chatbots

Any service that makes an AI chatbot available to users would have to clearly disclose that it is an artificial intelligence chatbot.

Restrictions on tampering with labels

The bill would make it unlawful to knowingly help hide, fake, remove, or falsify required AI disclosures. It would also ban making or selling tools or services mainly designed to strip off AI labels or add fake labels in order to mislead people about where content came from.

Enforcement

The Federal Trade Commission would enforce these requirements as unfair or deceptive trade practices. The Attorney General could also bring civil actions, and state attorneys general could sue as well. In some cases, private lawsuits could be brought by AI providers or covered online platforms that are harmed by violations.

Courts could issue injunctions, impound devices or products used in violations, award damages, and order modification or destruction of offending products. The bill also sets statutory damages for violations of the anti-circumvention rules, with higher penalties for repeated violations.

Exceptions and limits

The bill would not apply to AI-generated content used only for internal research and development that is not intended for public release or commercial deployment.

It would also include exceptions for certain activities, including:

• good-faith research on label robustness and detection tools;
• lawfully authorized law enforcement, intelligence, and related government activities;
• some activities by nonprofit libraries, archives, educational institutions, and public broadcasting entities, with limits for commercial or deceptive conduct.

Working group and standards

The bill would create a federal working group led by the National Institute of Standards and Technology, in coordination with the FTC, to develop technical standards and best practices for labeling, detection, interoperability, and enforcement. The group would include government agencies, AI developers, platforms, researchers, media organizations, privacy advocates, accessibility experts, and others. It would publish standards within a year, then submit a report to Congress. The group would later sunset, though it could be reconvened.

Relevant Companies

• GOOGL / GOOG — Alphabet’s Google products, including Search and YouTube, could need to detect, display, and preserve AI disclosures on covered content.
• META — Meta’s social platforms could need to label and preserve disclosures for AI-generated content shared by users.
• MSFT — Microsoft’s AI services and Bing-related products could be affected by labeling, provenance, and detection requirements.
• AMZN — Amazon-owned platforms that host user content, such as Twitch, could face disclosure and platform-compliance obligations if covered.
• SNAP — Snapchat could need to display and maintain AI disclosures for eligible content shared on the platform.
• PINS — Pinterest could be affected if it displays covered AI-generated images or similar content.
• PYPL — Payments-adjacent relevance is limited, but any tools or services involved in distributing content provenance or detection technologies could be indirectly affected.