Adobe clarifies that it will not train AI models from user content

Adobe has made changes to its terms of service agreement to state that it will not employ user content stored locally or in the cloud to train AI. The section about Adobe’s access to user content in the updated terms of service has added many categories and one talks specifically about generative AI. The terms clarify that it will not use Local or Cloud Connect to train generative AI except when the content is submitted to the Adobe Stock marketplace, in which case Adobe can utilise it to train Adobe Firefly.

This comes after Adobe faced outrage online over their terms of service update earlier this month. Adobe’s previous notification of new terms said that it might access the content through both automated and manual methods. Users interpreted the notification to mean that Adobe was allowing itself to freely access and utilise the work of users to train AI models. Users were forced to agree to the new terms to continue using the apps and services.

Adobe has also clarified that Adobe is not going through user content made under non-disclosure agreement (NDA), and has said that it does not scan or review work stored locally on the user’s device. Adobe has added that it will only scan content that is uploaded to the cloud to make sure that it is not holding any illegal or abusive content. And the content will be subject to human review if it is reported as illegal or flagged. Content will also be subjected to human review if the user opts in to beta, prerelease and Adobe’s product improvement programme or prerelease beta. Adobe has said “If you choose to share your Cloud Content with others using our Software and Services, we may automatically review this shared Cloud Content to flag abusive behavior (such as spam or phishing).”

Adobe shifted to a subscription only model in 2012.