OpenAI has announced that it will protect its business customers against copyright infringement lawsuits, following the lead of Microsoft, Google and Amazon, who already offer customers such protection in relation to their generative AI products.
The offer, dubbed the “Copyright Shield” by OpenAI, will apply to the generally available features of ChatGPT Enterprise, the paid-for business tier of ChatGPT, and to developers using ChatGPT’s application programming interface. It will not, however, apply to the free versions of ChatGPT.
Why has OpenAI announced this?
This announcement from OpenAI comes in the wake of increasing concerns about the huge amounts of copyright-protected works that ChatGPT and other AI-powered chatbots and generators are trained on.
As we’ve previously flagged on The Lens, those training processes might infringe copyright and other IP rights in the underlying training materials (see e.g. here, here and here) and a number of claims have been brought against generative AI developers in the US and the UK to test this. That includes claims against OpenAI themselves, who have been on the receiving end of various lawsuits in the US from a number of well-known authors including John Grisham, George R.R Martin, Michael Chabon, Jodi Picoult and Sarah Silverman.
In addition to the risk of IP infringement through training these tools, there is also a risk that users may receive an output that reproduces all or part of an underlying work that was used to train the system. As a result, those outputs (and any use of them) might also infringe the IP rights in the underlying training materials.
So far, we haven’t really seen claims of this kind being brought directly against users of ChatGPT or other generative AI tools. But it’s fair to say that people are becoming more and more aware of the potential risks involved and so OpenAI will be hoping that this offer seeks to allay some of those concerns and reassures its users that they will be protected if such a claim does arise.
What does OpenAI’s offer to protect users from copyright claims mean in practice?
The announcement that OpenAI will defend customers against legal claims of copyright infringement and indemnify them has obvious benefits for such users. However, it will not necessarily solve all issues for customers if a copyright holder does successfully claim that a ChatGPT output the customer has used infringes that copyright holder’s rights. For example, absent a licence, the user will still be prevented from using the relevant output. So in some ways, the risk still sits with the user.
There are also several limitations in the scope of the indemnity itself. For example, it looks like it won’t apply if: the user “knew or should have known” the output was infringing or likely to infringe; the user doesn’t use certain filtering and safety features provided by OpenAI; the output was modified or used with products or services not provided by OpenAI; or if the user didn’t have the right to use the relevant inputs (e.g. prompts) for generating the output. The indemnity also expressly excludes claims relating to trade mark infringement. That said, although trade mark infringement may be a risk (and indeed the claims made by Getty Images against Stability AI in the UK include claims of trade mark infringement relating to the use of Getty Images’ watermark), it will usually be easier for a user to spot and avoid potential trade mark infringement issues, as compared to copyright infringement.
More fundamentally, however, it appears that the indemnity is subject to a cap equal to the amount of fees paid by the user in the previous 12 months. If that’s right, the user may still be left (significantly) out of pocket.
From OpenAI’s perspective, this is not just about allaying customer concerns and thereby increasing user numbers. Offering to indemnify its customers in this way will enable them to take control of the defence of any such claims. That is really important as a finding that any ChatGPT output infringes copyright could have potentially huge systemic consequences for OpenAI’s business, and could be highly damaging from a brand perspective. Taking a somewhat cynical approach, one might argue that offering this new indemnity is a calculated step by OpenAI that arguably benefits them as much as (if not more than) its users.
What remains eminently clear is that developments in this space are not slowing down (indeed, at its developer conference OpenAI also announced a number of new features and other developments, including the unveiling of its new, more capable, cheaper GPT-4 Turbo model). With the use of generative AI tools like ChatGPT only on the rise (Altman has said it now has 100 million weekly active users), we can expect the debate over how to balance the rights of AI developers, users and IP rights holders to continue to rage on.