Does Microsoft Copilot's copyright indemnity protect users?

Generative AI tools from Microsoft and Adobe offer 'commitments' to protect users from infringement claims. Marisa Woutersen unpacks what they offer and when they might kick in.

Microsoft is “bullish” on the benefits of artificial intelligence (AI), but also “clear-eyed” about the challenges and risks associated with it, according to the company’s deputy general counsel Antony Cook.

Undoubtedly, the risks have come into sharp focus lately. In December, the New York Times sued OpenAI and Microsoft over allegedly infringing its IP in the training of its AI models—claiming that the pair had used this work to develop and commercialise their generative AI products without obtaining its authorisation.

A surge in copyright infringement cases linked to generative AI has prompted companies such as Microsoft, Adobe, and Amazon to take drastic measures.

The trio have each pledged to take responsibility for any potential legal risks and damages arising from copyright infringement claims incurred by users.

For Microsoft, this takes the form of a customer copyright commitment.

“It is our responsibility to help manage these risks by listening to and working with others in the tech sector, authors and artists and their representatives, government officials, the academic community, and civil society,” as Cook tells WIPR in a fresh interview since the launch of the pledge.

With the catchy tagline: “Use Microsoft Copilot services with confidence”, the ‘benefit’ extends the company’s IP indemnity support to commercial users of its AI Copilot tools.

But what exactly do these unusual indemnities cover and are they sufficient to allay the fears of users?

Unpacking Microsoft's indemnity clause

Microsoft introduced the Copilot Copyright Commitment to give “customers greater confidence” in September 2023.

Due to customer interest, the commitment was extended to include Microsoft’s Azure OpenAI Service in November 2023.

The indemnity states that if a third party sues a commercial customer for copyright infringement for using Microsoft’s Copilots or Azure OpenAI, or for the output they generate.

What’s more, Microsoft “will defend the customer and pay the amount of any adverse judgments or settlements that result” from the lawsuit.

“This benefit applies as long as the customer uses the guardrails and content filters we have built into our products,” explains Cook.

For instance, Microsoft has applied content filters to these products and services aimed at “preventing [users] from producing copyrighted materials,” he adds.

Cook emphasises that “customers are required to use these protections to be eligible for the customer copyright commitment”.

For the Azure OpenAI Service, Microsoft also offers documentation and tools that support the responsible use of AI and reduce risks of infringing copyrighted content.

Scrutinising AI indemnity terms

According to Ellen Keenan O’Malley, senior associate at EIP, the increasing number of lawsuits hitting the headlines is “inevitably going to make consumers nervous and question their liability risk if using any of its output data”.

She suggests it was “only natural” for AI creators to be introducing terms that “appear to address worries, demonstrating they stand boldly behind their technology.”

Ian DiBernardo, partner and chair of the IP litigation practice group at Brown Rudnick, agrees, pointing out that the move of generative AI platforms to provide indemnification for copyright infringement is “not surprising, with concerns from the IP sector focusing on “both the inputs and outputs”.

However, when you look closer at these terms, questions arise, cautions Keenan O’ Malley.

With regard to Microsoft's commitment to indemnify its commercial customers against copyright claims, she points out that “this is not an automatic right, and this is where users need to be careful”.

Customers must ensure they are—and have been at all times—compliant with the technical mitigation measures and best practices set out in the Azure OpenAI Service documentation.

Further, they must have also suffered an adverse judgment for copyright infringement for using the outputs generated by Azure OpenAI Service.

"The customer has to not only deal with the expensive and time-consuming litigation but also prove they infringed copyright but followed all of Microsoft's technical rules during the alleged infringement period,” explains Keenan O’Malley.

"Only then can the customer recover their losses," she adds.

She also queried whether this would then cover all legal costs or only the damages.

This is “not clear” and there is potential to never find out as the issue could be resolved before reaching the court to avoid fees and reputational damage, begging the question of the value of such indemnity,” argues Keenan O’Malley.

T&Cs comparison: Adobe vs Microsoft

In Adobe’s indemnity commitment, it offers the opportunity to obtain an IP indemnity for use of its content generated by certain Firefly-powered workflows.

However, it doesn’t seem to be an automatic right.

“The Adobe announcements don’t make it clear what the customer needs to do to get the IP indemnity—it may be that they have to pay an additional fee to benefit from it,” says Keenan O’Malley.

DiBernardo explains: “As with most commercial technology indemnification provisions, the generative AI providers include “carve-out and conditions to coverage.”

This means that while the provider can control the training of the model, and limit infringement liability, users can still increase their risk by applying prompts.

He explains that companies have taken steps to offset these potential risks.

“Adobe, for example, reduced its risk by training its initial Firefly model using Adobe Stock images, openly licensed content and other public domain content,” says DiBernardo.

“Microsoft mitigates exposure by providing conditions to indemnification coverage, from requiring meta prompts against infringement, to technical limits on inputs, and testing to detect third party content.

“Time will tell whether claims for indemnification will be made and whether, and to what extent, there may be disputes over coverage,” he adds.

What should customers consider?

Emma Varty, IP lawyer at Fieldfisher, explains that it will be the AI provider who will be liable for infringement taking place during an AI model's training phase—rather than the customer.

“Customers who want to use generative AI tools in their business should consider providers who guarantee proper licensing of training data, as this helps uphold ethical AI practices and supports the rights of data owners,” says Varty.

These customers should assess how AI providers handle both the training data and the content generated in their terms and conditions.

Other IP rights may also be infringed during a generative AI model's life cycle, so customers should also consider “whether indemnities limited to copyright infringement are sufficient,” she explains.

Keenan O’Malley suggests customers read all indemnities carefully to understand precisely what it covers and confirm it meets their indemnity needs, as well as “ensuring whether any further technical requirements internally need to be put in place” before use.

DiBernardo also urges users of generative AI platforms to also consider updating policies to reflect the conditions to receiving indemnification.

Amazon and Microsoft’s terms specifically require that the user retain and provide sufficient records to evaluate eligibility for indemnification.

“If there is a claim, users will need to establish eligibility, and having a record of compliance with the terms of use and satisfaction of any conditions will be important. The who, what, and when may matter,” concludes DiBernardo.

Did you enjoy reading this story?  Sign up to our free daily newsletters and get stories sent like this straight to your inbox.