The CLEAR Act requires companies using copyrighted works to train generative AI models to notify the U.S. Copyright Office and face penalties for non-compliance.
Adam Schiff
Senator
CA
The CLEAR Act establishes new transparency requirements for companies developing generative AI models. It mandates that developers must notify the U.S. Copyright Office about copyrighted works used in their training datasets before releasing or commercially using the model. This legislation also creates a public database of these notices and establishes civil penalties for non-compliance.
AI Companies Face New Transparency Rules: CLEAR Act Mandates Public Disclosure of Copyrighted Training Data Starting in 2024
The CLEAR Act is stepping in to pull back the curtain on how artificial intelligence is built. Starting 180 days after it becomes law, any person or company using copyrighted works to train or release a generative AI model—those systems that churn out text, images, or music—must file a formal notice with the U.S. Copyright Office. These notices aren't just internal paperwork; they will be published in a searchable online database for the whole world to see. For new models, this filing has to happen at least 30 days before the tech hits the market. For the AI tools you’re already using, companies have a 30-day window to come clean once the government sets the final rules. This is a massive shift toward transparency in a field that has largely operated as a 'black box.'
The Digital Paper Trail Under Section 2, companies have to provide a 'sufficiently detailed summary' of every registered copyrighted work they used to train their algorithms. Think of it like a nutrition label for your favorite AI chatbot. If you’re a freelance illustrator or a songwriter who has registered your work with the Copyright Office, this database is your new best friend. It allows you to see if a tech giant used your portfolio to teach their AI how to mimic your style. However, there’s a catch: the bill only covers 'copyrighted works' that are officially registered. If you have a blog or a collection of photos that are technically protected by copyright but aren’t registered with the government, those might still be used without appearing in these mandatory reports.
The Cost of Keeping Secrets This bill has real teeth when it comes to enforcement. If a company 'forgets' to file a notice, a copyright owner can take them to federal court. Under the enforcement provisions, judges can slap companies with civil penalties of at least $5,000 for every single instance where a notice wasn't filed. While there is a $2.5 million annual cap on these fines, that’s still a significant chunk of change for a startup. For a small business owner developing a niche AI tool, the administrative burden of cataloging thousands of data points could be a major headache. On the flip side, the bill allows courts to issue injunctions—basically a 'stop' order—that could pull an AI model offline until the company complies with the reporting rules.
Defining the Details While the goal is clarity, there is some 'Level Medium' vagueness in the phrasing. The bill requires a 'sufficiently detailed summary,' but it doesn’t explicitly say what that looks like. Does the company have to list every individual book title, or can they just say '10,000 mystery novels'? The Register of Copyrights has 180 days to figure out those specifics. For the tech workers coding these models and the creators whose work fuels them, the devil will be in those upcoming regulations. This law doesn't stop AI from using copyrighted data, but it ensures that if they do, they have to leave a public trail that creators can follow.