Content Origin Protection and Integrity from Edited and Deepfaked Media Act of 2025
- Sponsor
- Sen. Cantwell, Maria [D-WA]
- Committees
- Commerce, Science, and Transportation Committee (primary)
- Last action
- Apr 9, 2025
Bottom line
This bill aims to increase transparency and protect original content creators in the age of generative artificial intelligence by mandating provenance information and restricting unauthorized AI training data use.
What it actually does
This bill establishes a framework for transparency regarding synthetic and synthetically-modified digital content, including deepfakes. It mandates the development of standards for content provenance information and detection tools, requires commercial creators of AI content tools to enable provenance information, and prohibits the removal of such information or the non-consensual use of copyrighted content with provenance data for AI training.
Proponents argue
Supporters argue the bill is crucial for combating misinformation and disinformation spread through AI-generated content, protecting artists and creators whose work is used without consent to train AI systems, and fostering trust in digital media. They believe it will catalyze innovation in responsible AI development and establish the U.S. as a leader in this field.
Opponents contend
Critics might contend that the requirements for content provenance information and restrictions on AI training data could stifle innovation in the AI industry, impose significant compliance burdens on developers and platforms, and be technically challenging or impossible to implement effectively at scale. Concerns could also be raised about potential impacts on free speech or the open-source AI community.
The bill is relatively short and its core provisions are understandable, making it accessible for an informed reader to evaluate within a reasonable timeframe.
Section 4(a)
Establishment of Public-Private Partnership for AI Content Standards
This section requires the Under Secretary of Commerce for Standards and Technology (NIST) to create a public-private partnership. This partnership will facilitate the development of voluntary, consensus-based standards and best practices for watermarking, content provenance information, and the detection of synthetic and synthetically-modified content across various media types (images, audio, video, text). It also includes developing metrics for evaluation tools and establishing grand challenges for detection and cybersecurity countermeasures.
Supporters argue
Supporters argue that establishing clear, consensus-based standards is essential for the responsible development of AI, ensuring interoperability, and building public trust in digital content. This proactive approach fosters innovation while mitigating risks.
Critics contend
Some might argue that government-mandated or facilitated standards could be premature in a rapidly evolving field, potentially stifling innovation or favoring certain technologies over others. Concerns about the practicality of achieving true consensus could also be raised.
Tradeoffs
Balances the need for rapid technological advancement with the desire for responsible and ethical development, potentially trading some speed for greater trust and interoperability.
Section 5(a) and 5(b)
NIST Research Program and Public Education Campaign
This section directs NIST to conduct research and development into the robustness and efficacy of technologies for synthetic content detection, watermarking, and content provenance, as well as cybersecurity protections against tampering. Additionally, within one year, NIST must launch a public education campaign about synthetic content (including deepfakes), watermarking, and content provenance information.
Section 6(a)(1) and 6(a)(2)
Mandate for AI Content Tool Provenance Capabilities
Starting two years after enactment, any person commercially making available tools primarily used to create synthetic or synthetically-modified content must provide users the ability to include machine-readable content provenance information indicating the content's synthetic nature. Similarly, tools for creating or substantially modifying "covered content" (copyrightable works) must offer users the ability to include provenance information. In both cases, reasonable security measures must be established to prevent easy removal or alteration of this information.
Section 6(b)(1) and 6(b)(2)
Prohibition on Tampering with Content Provenance Information
This section makes it unlawful for any person to knowingly remove, alter, tamper with, or disable content provenance information in furtherance of an unfair or deceptive act. Covered platforms are specifically prohibited from removing, altering, tampering with, or separating provenance information from content, with an exception for security research.
Section 6(c)
Restriction on Using Provenance-Tagged Content for AI Training Without Consent
This section makes it unlawful for any person, for a commercial purpose, to knowingly use "covered content" (copyrightable works) that has attached provenance information, or from which provenance information was illegally removed, to train an AI system or generate synthetic content, unless they obtain express, informed consent from the content owner and comply with any terms of use, including compensation.
Section 7(a), (b), (c)
Enforcement by FTC, States, and Private Parties
This section outlines the enforcement mechanisms. Violations are treated as unfair or deceptive acts under the Federal Trade Commission Act, allowing the FTC to enforce the Act. State attorneys general can also bring civil actions as parens patriae. Additionally, private parties who own covered content with provenance information can bring civil actions against those who tamper with it or use it without consent for AI training, seeking injunctive relief, compensatory damages, and attorney's fees. A 4-year statute of limitations applies to private actions.
Defines "covered platform" as a website, internet application, or mobile application available in the U.S. (including social networking, video sharing, search engines, content aggregators) that either generates at least $50,000,000 in annual revenue OR had at least 25,000,000 monthly active users for at least 3 of the preceding 12 months.
Section 3(6)
Why it matters:Standard legislative practice to define terms upfront, but the specific thresholds are highly consequential.
Case for: Proponents argue these thresholds target large, influential platforms that have the greatest impact on content dissemination and the resources to comply, while exempting smaller startups.
Case against: Critics might argue that these thresholds are arbitrary or could be easily circumvented, or that even smaller platforms can have significant impact and should also be held accountable.
Estimated impact: Limits the scope of platform liability and requirements to major tech companies, affecting potentially hundreds of millions of users.
The requirements for providing provenance information in AI tools and the prohibition on non-consensual use of covered content for AI training only apply when the activity is for a "commercial purpose."
Section 6(a)(1), 6(a)(2), 6(c)
Why it matters:Standard practice to limit the scope of regulation to commercial activities, but it has significant implications for non-commercial or academic use.
Case for: Proponents argue this protects academic research, open-source development, and non-profit initiatives from undue burden, focusing regulation on profit-driven activities where the risk of exploitation and deception is highest.
Case against: Critics might argue that non-commercial AI development can still produce harmful synthetic content or exploit copyrighted works, and that this carve-out creates a loophole.
Estimated impact: Exempts a significant portion of AI development and content creation from the bill's mandates, particularly in research and open-source communities.