A Creative Cloud setting has prompted questions about whether Adobe uses customers’ photos to train AI. The answer depends on what kind of analysis and training are meant: Adobe says it may analyze content to improve products, while also saying it does not use Creative Cloud account data to train its experimental Generative AI features.
What the setting says
The setting is called “content analysis.” It says Adobe may analyze content using techniques such as machine learning, including for pattern recognition, to develop and improve its products and services. A developer at Krita recently drew attention to the setting in their Adobe Creative Cloud account, and the article says users had been opted into the program.
That wording raises a reasonable question: does analyzing content mean using photos to train AI? The article distinguishes between machine learning used to analyze or organize content and training machine learning systems. Those activities can be related, but the setting itself does not say that customer images are used to train generative models.
Analysis can support product improvements
Adobe’s explanation gives examples of machine learning features that can help people organize and edit images. In Lightroom, object recognition can automatically tag photos of a dog or cat. In Photoshop, machine learning can automatically correct an image’s perspective.
Analysis could also help Adobe understand broad patterns in how people use its products, such as whether they edit images of people or landscapes. That kind of information could inform product decisions and priorities. The source does not say that this is the only use of content analysis, but it illustrates a purpose distinct from training a generative AI system.
There is also a difference between analyzing individual content and using aggregated content. Adobe’s documentation says that, for product improvement and development, it first aggregates content with other content, then uses the aggregated content to train algorithms and improve products and services. So the documentation does describe training algorithms with aggregated content.
Adobe’s statement on Generative AI
Adobe told PetaPixel that “Adobe does not use any data stored on customers’ Creative Cloud accounts to train its experimental Generative AI features.” That statement addresses a specific category: experimental Generative AI features. It does not erase the separate documentation about using aggregated content to train algorithms for product improvement and development.
The distinction matters because “training algorithms” can sound like one single practice, even though the source describes different purposes. Adobe’s statement concerns experimental Generative AI; the documentation concerns product improvement and development. The article says the available wording leaves open the possibility that photos could, through some route, be used to train a generative AI, but does not establish that this is happening.
The source also notes that Adobe has a separate Adobe Photoshop Improvement Program, which is opt-in and documented. It raises the possibility of manual review in some circumstances, while treating that as a distinct privacy issue. It does not provide further details about those circumstances.
What Creative Cloud users can consider
For users deciding what to do, the key is to read the content analysis setting according to its stated scope and keep Adobe’s separate Generative AI statement in view. The article recommends opting out of this program and any others if privacy is a priority. It says users can do that through the privacy page when logged in.
That recommendation follows from the uncertainty the article describes: the setting permits analysis for product development, and Adobe’s documentation describes training algorithms with aggregated content. Adobe’s statement rules out using Creative Cloud account data to train its experimental Generative AI features, but the source says that does not settle every possible question about how content may be used. Reviewing the settings and opting out gives users a way to respond to that uncertainty.