On 10 April 2025 US senators reintroduced the Content Origin Protection and Integrity from Edited and Deepfaked Media Act (COPIED Act). Issued by the Senate Committee on Commerce, Science and Transportation, the bill carries forward the core approach of the earlier version.
Three keywords: mark, authenticate, detect
The COPIED Act’s mechanism has three parts: marking AI-generated content, authenticating it through content provenance information, and supporting the detection of unmarked content. Together they point to one goal: making the origin of content traceable.
This is close to the EU’s path on labelling AI-generated content. See the draft EU code of practice on AI content labelling and China’s synthetic content labelling regime. The instruments differ between jurisdictions, but all are trying to answer the same question: how machine-generated content can remain identifiable as it travels along the distribution chain.
A tilt towards content creators
The bill expressly protects groups such as journalists, actors, artists and songwriters, giving them stronger control over the use of their work for training or generation. That orientation echoes the US Copyright Office’s position on training data.
See the US Copyright Office report on generative AI training and our notes on protecting AI model parameters. What companies need to watch is not whether the bill passes, but that the burden of proving the lawful origin of training data is being reinforced again and again.
How it divides work with TAKE IT DOWN
The TAKE IT DOWN Act advanced in the same period focuses on the duty to remove non-consensual intimate imagery, while the COPIED Act focuses on the transparency of content provenance. One handles the rapid takedown of harmful content, the other the tracing of generated content, and together they form a combined response to deepfakes.
Three things export-facing companies can prepare
First, keep content provenance records upstream. Whichever jurisdiction ultimately applies, being able to show the origin of training data and generated content is the core evidence in a dispute.
Second, watch how platform-side duties flow downstream. Labelling and provenance authentication obligations often fall first on platforms and are then passed on by contract to model and tool providers. See our notes on the user verification duty for AI-generated content.
Third, treat labelling capability as a product feature. For a content product operated in the EU, the US and China at the same time, the labelling scheme needs to be designed once rather than patched jurisdiction by jurisdiction.
Legislative progress will rise and fall, but the direction of traceable content provenance will not reverse.
This article is based on material on the COPIED Act from the US Senate Committee on Commerce, Science and Transportation and public information and is for reference only; it does not constitute legal advice.