We’ve heard a lot recently about Big Tech indemnifying customers against generative AI Intellectual Property (IP) claims from the copyright industries.[1] But how do these indemnities work and is there a catch?
The Microsoft Customer Agreement (MCA)
As an example, Microsoft’s Customer Copyright Commitment has attracted a lot of interest. The start point here is the MCA, the umbrella agreement that calls down all Microsoft’s other terms.[2]
The MCA itself contains an indemnity where Microsoft commits to defend the customer against third party IP claims. This indemnity is hedged with a number of conditions, including:
- Microsoft is given prompt notice of the claim and controls its defence;
- the product must be paid for;
- it must be used within licence scope, unmodified and not combined with anything else; and
- Microsoft can terminate the licence and refund fees paid if it can’t settle the claim.
The Microsoft Product Terms
One set of terms that the MCA calls down is the Product Terms, and it’s here in the Gen AI Services section that the Customer Copyright Commitment lives.[3] This operates as an extension of the indemnity in the MCA to cover IP claims arising from the output of Microsoft Copilot or an Azure OpenAI service, in each case where five more conditions are met:
- the customer didn’t tamper with any Copilot or OpenAI safety systems;
- the customer has the necessary rights to the input;
- the customer didn’t use the output in circumstances when it knew, or should have known, that it was likely to infringe third party rights;
- the claim isn’t for trade mark infringement; and
- for Azure OpenAI, the customer must have implemented all mitigations required by the product documentation.
What the customer wants
With the rising tide of genAI infringement claims, IP indemnities are more important they look.
The rationale is that the customer says to the provider: ‘I want to buy your service, but you’re the only person who knows if you’ve infringed someone else’s rights, so please bear that risk and reimburse me for anything I have to pay if I’m on the wrong end of an infringement claim’.
But it’s never that simple and IP indemnities are complex and come festooned with conditions. And some of these are far reaching, so as the customer, it’s really a case of looking at the small print. Here you might think that some of the requirements are a bit tongue in cheek:
- what does using the product ‘uncombined’ mean? Surely the customer will always be using the product combined with something else?
- does the provider’s right to terminate the licence if it can’t fix the problem take away protection just when it’s needed most?
- when “should the customer have known” that a claim was “likely”?
What the provider is willing to commit to
And what does the provider say to all this? Basically, ‘yes, we want to give our customers protection against the risk of claims; but no, there’s no catch: we’re simply not offering an unconditional commitment, particularly where the customer isn’t paying a lot. And you can decide whether or not to use our product on this basis.’
So as gen AI scales, the scene is set for a major tussle between Big Tech on one side, the copyright industries on the other and the customer in the middle – a tussle where the lowly IP indemnity is set to have a big role.
[1] For example:
- Google: Shared fate: Protecting customers with generative AI indemnification, Neil Suggs (VP Legal, Google Cloud) and Phil Venables (VP, TI Security & CISO, Google Cloud), 13 October 2023
- AWS: AWS Customer Agreement, section 7 (Indemnification), 12 January 2024
- IBM: IBM Announces Availability of watsonx Granite Model Series, Client Protections for IBM watsonx Models, 28 September 2023
- Microsoft: Introducing the Microsoft Copilot Copyright Commitment, Microsoft, 8 September, 2023.
[2] Microsoft Customer Agreement
[3] Microsoft Product Terms > Universal License Terms >For Online Services
