When OpenAI Turned Copyright On Its Head
Recently, OpenAI, the creator of ChatGPT, announced that it would be launching a tool called Media Manager in 2025. It has been hailed as an “olive branch” to content owners. I’m calling foul on that.
According to reports, Media Manager will allow content owners to tell OpenAI what works they own and “express their preferences for how their content is used in AI.” Under this scheme, copyright owners will be able to identify their copyrighted works and select whether they want those works to be used in training AI systems. Sounds great. What’s the problem?
Copyright 101
The problem is that this is not how copyright works. As a refresher, Article I, Section 8, Clause 8 of the U.S. Constitution gave Congress the power to “promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.” Consistent with this power, the Copyright Act gave content creators the exclusive right to do and authorize the following:
to reproduce the copyrighted work;
to prepare derivative works based upon the copyrighted work;
to distribute copies of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending;
to display the copyrighted work publicly or
to publicly perform the copyrighted work.
With very limited exceptions (which arguably are not applicable in the training of AI models), someone may not use a copyrighted work without express authorization from the copyright owner. Usually, this authorization comes in the form of either (a) a license, where the copyright owner grants permission under specific terms, or (b) a release, where the copyright owner agrees that it will not sue the user for certain specified uses of their copyrighted work.
My Billion Dollar Idea
So how is this different from what OpenAI has proposed? Let’s say I have this billion dollar idea for a new Star Wars movie. I can go out and secure financing and produce my movie, right? Clearly not. What I’ve described would be a derivative work of the Star Wars films that came before it. The right to make derivative works is an exclusive right of the copyright owner, in this case Lucasfilm. So before I go ahead with my project, I need to go to Lucasfilm and secure a license from them.
Under the scenario painted by OpenAI, I should be able to make my film because Lucasfilm has not notified me specifically that they don’t want me to make derivative works of their copyrighted content.
Let me say it again – THIS IS NOT HOW COPYRIGHT WORKS.
OpenAI is potentially in a bind because they went ahead and copied all (or nearly all) of the data available on the internet. As described above, copying is one of those pesky exclusive rights that are granted to copyright owners. Because of its unlicensed use of copyrighted works, OpenAI is being sued left and right by copyright owners.
So now they are saying that IF copyright owners identify their works via the Media Manager tool, and IF the copyright owner says that they don’t want their work to be used to train AI models, then OpenAI will respect the wishes of the copyright owner.
Inverting The Copyright Paradigm
Copyright law, as written, requires a user to go to a copyright owner and secure a license (or release) to use their copyrighted work. Copyright owners don’t need to tell anyone they have to do this -- it is codified in law. If a work is not yours, then you need to find out who owns it. If it is protected under copyright, then the obligation is on the potential user of that work to go out and get the license.
OpenAI’s proposed solution puts the onus on the copyright owner. What if that copyright owner owns thousands of copyrights? Will they have to register thousands of works on the Media Manager tool? And let’s remember that OpenAI’s ChatGPT is only one of an ever-increasing number of AI models out there. Will a copyright owner now have to register their works with OpenAI, Google, Meta, Anthropic and every other platform? This is suddenly becoming a lot of work. For something that copyright law already says is not the copyright owners’ responsibility. Today, once a copyright vests in a given work, the copyright owner is done. Any subsequent use of an exclusive right granted by copyright law is the responsibility of the potential user.
By creating an “opt out” system instead of an “opt in” system, OpenAI has flipped copyright law on its head. It is attempting to put all of the onus on copyright owners rather than on potential users, where it belongs.
More Questions Than Answers
Since OpenAI has already copied and ingested billions (trillions?) of copyrighted works without permission, this “opt out” system may be the best that it can offer. It is trying to close the proverbial barn door after the horse has escaped. But it is unclear thus far whether they are saying that they will go so far as to cause ChatGPT to “unlearn” whatever it learned from your copyrighted work. Even if they say that they will, should we believe them? Will they really go and scrub works entirely from their model? Or will they leave the work in the database, reaping the benefits of training on your work, but simply suppress the system from producing any results that display the copyrighted work in question? Even if they let copyright owners audit their system, would anyone know where or how to check this? If they suppress the system from producing any results that display the copyrighted work, is this a sufficient remedy to copyright owners?
This news is only a week old so there is still much we don’t know about OpenAI’s plans. We should view this with a lot of skepticism. A single (but powerful) company should not be permitted to rewrite copyright law in front of our very eyes. Will copyright owners buy into this approach? And will other countries around the world follow suit? Before we thank OpenAI for this “olive branch”, we should make sure the wood isn’t rotten.
Notes:
