Microsoft is enhancing its policy to safeguard commercial clients from copyright infringement lawsuits linked to the use of generative AI — but with several important conditions.
During the Ignite conference today, Microsoft announced that customers who license the Azure OpenAI Service, its fully managed platform that incorporates governance layers over OpenAI’s models, will receive support and compensation from Microsoft for any "adverse judgments" related to copyright infringement if they are sued while using the Azure OpenAI Service or its outputs.
Generative AI models like ChatGPT and DALL-E 3 are trained on vast datasets that include millions of e-books, artworks, emails, songs, audio snippets, and voice recordings, mostly sourced from public websites. While some training data is in the public domain, other aspects may require citation or entail specific compensation due to licensing agreements. The legality of utilizing data without permission is currently being debated in courts. A key concern for generative AI users is the risk of generating outputs that closely replicate elements of the training data.
Microsoft's new policy won't automatically apply to every customer using the Azure OpenAI Service. To qualify for these enhanced protections, subscribers must implement "technical measures" and adhere to specific documentation requirements designed to reduce the possibility of producing infringing content with OpenAI’s models.
Tech media reached out to Microsoft for more information on these technical measures, but the company did not provide additional details prior to this morning's announcement. It remains uncertain whether these protections apply to Azure OpenAI Service products currently in preview, such as GPT-4 Turbo with Vision, or if Microsoft offers indemnity against claims related to training data used by customers for fine-tuning OpenAI models. We sought clarification on these points.
Later in the day, a Microsoft spokesperson confirmed via email that the policy includes all products in paid preview and covers Microsoft’s training data — but not the training data utilized by customers.
This new policy follows Microsoft’s September announcement, in which it committed to cover legal damages for customers of select AI products who face copyright infringement lawsuits. Like the Azure OpenAI Service protections, customers must utilize the “guardrails and content filters” embedded in Microsoft's AI offerings to maintain their coverage.
Coincidentally, OpenAI recently announced plans to cover legal fees for customers facing IP-related lawsuits over work produced using its tools. Microsoft’s updated Azure OpenAI Service protections seem to extend this initiative.
Beyond indemnity measures, an interim solution to the regurgitation issue involves allowing content creators to opt out of having their data included in generative model training sets or to ensure these creators receive some form of credit and compensation. OpenAI has indicated it will explore this possibility with upcoming text-to-image models, potentially succeeding DALL-E 3.
Conversely, Microsoft has not committed to such opt-out or compensation strategies. However, the company has developed technology designed to help “identify when [AI] models generate material that utilizes third-party intellectual property and content.” This feature is included in the preview phase of Microsoft’s Azure AI Content Safety tool.
We sought further details about how this IP-identifying technology operates, but Microsoft only referred us to a general blog post. We will continue to monitor developments at Ignite for more insights.
Ranking
