Microsoft is expanding its policy to protect commercial customers from copyright infringement lawsuits arising from the use of genetic artificial intelligence — but with a caveat (or several).
Today during Ignite, Microsoft said that customers who license Azure OpenAI Service, the company’s fully managed service that adds layers of governance on top of OpenAI models, can expect to be defended—and indemnified—by Microsoft for any “adverse judgments” if they are sued. for copyright infringement when using the Azure OpenAI service or the results it generates.
AI models like ChatGPT and DALL-E 3 are trained on millions to billions of e-books, artwork, emails, songs, audio clips, voice recordings, and more, most of which come from public websites. While some of this training data is public, some is not — or covered by a license that requires reporting or specific forms of compensation.
The legality of training vendors on unlicensed data is another issue fracturing the courts. But what could land genetic artificial intelligence users the problem is regression, or when a production model spits out a mirror copy of a training example.
Microsoft’s expanded policy will not apply by default to every customer of the Azure OpenAI service. To be eligible for the new protections, subscribers are prepared to implement “technical measures” and comply with certain documents to mitigate the risk of creating infringing content using OpenAI’s models.
TechCrunch asked Microsoft for more details on these measures, but the company declined to provide details ahead of this morning’s announcement.
It’s also unclear whether the protections extend to Azure OpenAI service products in preview, such as GPT-4 Turbo with Vision, and whether Microsoft offers compensation against claims made about training data used by customers to improve OpenAI models. We asked for clarification.
Late this afternoon, a Microsoft representative told TechCrunch via email that the policy applies to all products in paid preview and Microsoft’s — but not a customer’s — training data.
The new policy follows Microsoft’s announcement in September that it will pay legal damages on behalf of customers who use some — but not all — of its artificial intelligence products if they are sued for copyright infringement. As with Azure OpenAI service protections, customers must use the “guardrails and content filters” built into Microsoft’s AI offerings to maintain coverage.
Perhaps not coincidentally, OpenAI recently said it would begin paying legal fees incurred by customers facing lawsuits over IP claims against work created by OpenAI tools. Microsoft’s new Azure OpenAI service protections appear to be an extension of this.
Compensation policies aside, a partial solution to the regression problem is to allow content creators to remove their data from creator model training datasets — or to give those creators some form of credit and compensation. OpenAI said it will explore this with future text-to-image models, perhaps as a follow-up to DALL-E 3.
Microsoft, by contrast, is not committed to any exclusion or compensation programs. But the company he’s got has developed a technology it claims can help “determine when [AI] models create material that leverages third-party intellectual property and content.” A new feature in Microsoft’s Azure AI Content Safety tool is available in preview.
We asked for background on how IP identification technology works, but Microsoft declined — simply pointing to a high-level blog post. We’ll keep our eyes open for more details at Ignite.
