A significant number of legacy software solutions are now incorporating generative artificial intelligence (GenAI), and most new software solutions have some form of GenAI capabilities. This is true across the majority of, if not all, industries and, as such, it is not surprising that we are seeing a large increase in GenAI-related queries from businesses that use, or are procuring, software.
These queries vary from liability, regulatory compliance, vendor due diligence, and bias issues through IP rights and indemnities. We have covered a number of these potential issues in some of our previous blog posts on artificial intelligence.
In this blog, we look at the relationship between GenAI and some typical intellectual property (IP) provisions in software supply agreements.
IP Ownership Provisions
All supply agreements that have some element of software supply, in addition to appropriate license/rights-of-access provisions, should include IP ownership provisions making it clear which party owns
- the software;
- the data to be input into the software;
- the content provided by the software; and
- any outputs/deliverables obtained from the software.
The exact ownership structure will depend on the specifics of the deal agreed by the parties, but, as the IP in question will generally be copyright created by a natural or legal person (a company), global laws generally provide clarity as to initial ownership of such IP, with such ownership capable of being varied or later transferred, in each case by way of agreement between parties.
When it comes to the use of GenAI, such certainty may be lost.
As with the development of most technologies over the decades, the law is generally lagging behind the development of GenAI solutions, with IP ownership legislation currently differing across the world in respect of the ownership of GenAI outputs.
In some jurisdictions, ownership is possible (this would likely be the creator of the GenAI solution) and in others, the law just does not address the issue, and therefore it is likely that the outputs created would be public materials without legal protection.
In light of this, it is important that contracts relating to the use of GenAI and its outputs address the ownership/licensing of such GenAI outputs in order to document the agreement of the parties in the absence of legislative protections. As for the terms of such ownership, there is no one-size-fits-all solution to this; it will depend on the details of the specific transaction/solution, including the commercials, leverage of the parties, and the use case(s) of the outputs.
The IP ownership structures in relation to GenAI solutions can be complex, but the main issues include: which party owns the inputs (the prompts/queries), which party owns the outputs of the GenAI, and—irrespective of ownership—what each party can do with the inputs and the outputs (i.e., the license rights provided by the owner). The parties will need to give careful thought to these issues and draft IP provisions to reflect their specific agreement.
IP Indemnity Provisions
Another critically important provision for software-related supply agreements is the IP indemnity provided by the supplier of the software solution.
These IP indemnities have become standard in such agreements, so much so that they sometimes are not given the attention that they deserve, becoming quasi-boilerplate provisions.
With the rise of GenAI solutions, the spotlight has now returned to IP indemnities and things are moving rapidly.
For a period of time, suppliers of GenAI solutions were providing little-to-no protection to the users of their solutions, including in respect of the outputs of the GenAI infringing IP rights. Given that a number of the early solutions were being provided on a pilot/beta/free basis, this position made some sense. However, now that the market is developing, with GenAI solutions starting to be monetized and used for higher-risk activities, the position has shifted, and we are seeing demand from customers to receive appropriate indemnity coverage and suppliers sometimes willing to provide such coverage. This is particularly the case since some of the hyperscalers announced last year that they would provide standard indemnities in relation to some of their GenAI solutions.
While GenAI output indemnities from suppliers are now becoming more common, we have seen the scope of such indemnities from suppliers vary greatly. The inclusion of an indemnity and its scope depends on the solution that the supplier is offering. For example:
- If outputs are based on the supplier’s own large language model (LLM)/training data, suppliers may be more confident providing a GenAI output IP indemnity that is more akin to a traditional IP indemnity
- If outputs are based on data provided by the customer, it is unlikely that the supplier will offer an IP indemnity; in fact, it is likely that the supplier will require an indemnity from the customer in respect of IP infringements resulting from the customer’s data
- If the solution operates based on a third party’s solution/LLM, suppliers may try to resist providing a GenAI output IP indemnity or, failing that, only offer what can be backed off with the third-party provider (e.g., flowing up any protections that the supplier receives from the third party).
We expect that, as GenAI technology and the legislation/regulation of such technology matures further, indemnities will become more and more commonplace as risks can be more accurately assessed; however, at the moment, there is little consistency, with suppliers still often only able to offer limited protections to customers due to the design of their solution and the commercials of the particular transaction.
It is worth noting that, even where IP indemnities for GenAI outputs are provided, they may be subject to heavy carve outs and limitations/exclusions of liability and should be reviewed carefully to understand the likelihood they could be relied upon if an infringement issue actually arose. Parties may also want to consider if the solution/outputs can be split between GenAI and non-GenAI. If this is possible, then the IP indemnity could potentially also be split, with the first limb being a more traditional IP infringement indemnity and the second limb being a GenAI output specific IP indemnity, which may differ in scope. This may help mitigate the risk of watering down standard IP indemnity protections when GenAI is introduced as part of the solution.