Did Adobe Steal Your Photos to Train Their AI? We Asked Them in Person.

This past year, Adobe rolled out a terms-of-use update that shook the creative community. The way this update was worded made it appear as though you were giving them complete rights to your artwork just by using their services in any way. Users were rightfully scared, and the pitchforks came out. There have been some updates to the situation, and I recently went to Adobe MAX to speak with them directly.

So how did this all start again? It was a seemingly innocuous update to the terms of use—you know, the thing that most of us just click "yes" to agree to and move on without reading. However, those who did read it were shocked to see that the language used gave Adobe the rights to their content. The buzz over the internet—on YouTube videos, articles, and even in the mainstream media—was fierce. In response to this huge pushback, Adobe posted multiple blogs and even made a subsection of their site just to clarify and explain their terms. Additionally they updated them again to be more clear. So let's dive into the details.

Here is the terms of use update that shook the industry:

Solely for the purposes of operating or improving the Services and Software, you grant us a non-exclusive, worldwide, royalty-free sublicenseable, license, to use, reproduce, publicly display, distribute, modify, create derivative works based on, publicly perform, and translate the content. For example we may sublicense our right to the Content to our service providers or to other users to allow the services and software to operate as intended, such as enabling you to share photos with others.

Now you can see their latest updates, as of this article:

Adobe's explanation for why this change in the terms even occurred in the first place does make sense, but it looks like they could have done a better job wording it from the start. Their new explanation is that, in order to use these Generative AI Services, the AI does have to look at your content to perform its actions. For example, if you're having it remove distractions from a photograph in Photoshop, it does have to analyze the photograph. Their intention was that by agreeing to the terms of use, you're just giving them permission to access the specific content. Therefore, the terms just cover them legally by your agreeing to a statement regarding that permission.

Adobe has also published a new statement that they never assume ownership of customers' work.

The second point: Many users were, and still are, concerned about where the AI was and is getting its information to train these datasets. There are countless videos online still, as I write this, that claim Adobe uses user data.

How did Adobe train their generative AI models? Was it on customer content? Did Adobe just scrape the internet or use your photos to train its AI without you even knowing?

Adobe addresses this very directly and bluntly on their new frequently asked questions and clarification page. They state that:

Adobe does not train Firefly Gen AI models on customer content. Firefly generative AI models are trained on a dataset of licensed content, such as Adobe Stock, and public domain content where copyright has expired.

Public domain is content that exists where copyright has expired or does not apply. Laws vary by country. For example, in the United States, we look to the Copyright Act of 1909 (Public Law 60-349) which was signed by President Theodore Roosevelt.

So, just based on the statements, blogs, and other content that Adobe has put out publicly online, the clarification again is that they're not going to steal your photos, nor did they use customer content to train Generative AI in the first place.

Interviews at Adobe MAX

Aside from reading about these events and watching videos, I also went, as the saying goes, to "hear it straight from the horse's mouth." Adobe recently hosted a huge event called Adobe MAX in Miami, Florida. As press, I attended the event and met with Adobe staff, including those who directly work on these programs. Let's start at the beginning of the event.

First off, in the Adobe MAX keynote kicking off this extraordinarily large event, Adobe's Chair and Chief Executive Officer, Shantanu Narayen, took to the stage. In his opening keynote, one of the very first things he did was discuss Generative AI. His talk covered their feelings of responsibility toward the creative community and the implications of these new technologies for all of us. Mr. Narayen explained that Adobe is leading industry standards and at the forefront of what's going on. He looked at the many hundreds to thousands of people in the crowd and said in front of all of us that Adobe, again, is not training their generative AI on user data. I even double-checked and rewatched his talk to hear it again. In the recording of the opening keynote, he makes this statement at about 5 minutes and 30 seconds in. If this is a topic that you're interested in and you have an Adobe account or can find it online, I encourage you to watch this keynote. This way, you can see firsthand how Adobe is addressing these issues outside of just my own summary for you. While I strongly believe that they may have gone awry with the initial terms-of-use debacle, they're definitely putting effort into correcting misinformation and misinterpretation.

I'm press. I still had questions. So, while at MAX, I was fortunate to have scheduled interviews with Adobe staff where I could speak with them directly about Generative AI. In particular, I spoke with Stephen Nielsen, Senior Product Manager for Photoshop. Afterwards, I met with Rob Christiansen, the Director of Product Management for Adobe Lightroom and Adobe Camera Raw. They were both kind enough to sit down with me personally for solo interviews where we discussed Adobe's latest announcements and generative AI advancements.

It's important to note that, as human beings, there's a big difference between a giant multibillion-dollar company making statements and you sitting down face to face with someone on the team and talking about it. Stephen was warm and inviting, and I sensed no disingenuous intentions in his explanation and agreement that the clarifications are all correct. The AI never trained on user data, and they actually really care about the community. This sentiment was later backed up in my cross-check with Rob. They were both wonderful to talk with and very open and inviting. I even later saw them participating in photo walks and enjoying Adobe MAX. It definitely gave a different side that I want to share with you all: that these big companies have real people behind them, and at least in this case, those people do care. After MAX, several Adobe team members reached out, thanking me for joining them at the event and letting me know to contact them if I had any questions or feedback.

Caring about the community is a big statement.

For Context, a Look at Meta’s Terms

I want to make a sidebar here that, in writing this article, I did research not just on Adobe and what they're doing with training their AI models but also on what other big companies are doing. I wanted to get the temperature in the room to see where Adobe stands on the scale in comparison. So I went straight to the top, where most creatives share their work. This is not fluff to make Adobe look good, but it is factual that in Meta's terms, they state that if you post to Facebook, Instagram, or use their services, they are training on your public data. They aren't the only ones. There are so many companies—from social media to cloud services—that are point-blank unapologetically saying if you use their service, they are using your data. In most cases, there is no way to opt out of this unless you live in certain countries or regions. For example, with Meta in the EU, it's outlawed to not be able to have a choice, so EU users can supposedly opt out. As a United States-based user, this is what I see:

For example, if you use Instagram, you can go into your settings right now and see that they are very open about the fact that your content that you post publicly has been used to train their generative AI as well as for their open-source community. Their only clarification is that they only use public posts to train the AI. So your private content should still be safe.

Content Analysis and Product Improvement Programs

Well, back to Adobe. We've seen what they've clarified online, we've talked to them personally, so where do things stand? As an avid user and enjoyer of Lightroom, Photoshop, and Premiere Pro, I looked at my Adobe account. If you dig into your Data & Privacy settings, there's a page with a toggle switch about content analysis. This panel allows you to opt out of Adobe analyzing your content for Product Improvement Programs. It does state that it has nothing to do with training Generative AI models. The setting instead states that it has to do with machine learning to improve their products and services. When you click the "learn more" button, it does explain that it's not looking at the content on your computer, phone, or tablet. Instead, it may look at content processed by or stored on their servers. There isn't much clarity on what the analyzing means, but at least it's stated in a very legal way that it's not for Generative AI. I also did click the opt-out button. The fact that we can even opt out is promising. However, when you first look at the setting, it is already toggled on. In researching further, accounts that are business accounts or school accounts are automatically opted out, and if you have one of those accounts, you won't even see the toggle option. It's only personal accounts that are affected by this—just so you know.

My Opinion

I'm going to give you my personal opinion. I think that these artificial intelligence models are something so new, and they're now hitting the mainstream. I believe that these things are happening fast, and legal teams are trying to cover their clients—these big companies—by making very blanket statements so that way, in the next step or next iteration, these terms will cover whatever direction the river flows, so to speak. These legal terms of use are meant to cover them, not us. They're written by the company's side, so obviously that's who they're trying to protect. It's easy to get upset when something isn't handled the right way, and my personal belief is that the original terms of use were too broad. The wording made it seem as though you were giving up the rights to your work. I've written before about photo contests that I feel are just rights grabs, and this had that same kind of gross feeling to a lot of people. But I'm glad to see Adobe walk it back and understand that there's a middle ground to protecting themselves but also protecting us. As Adobe's Chair and Chief Executive Officer, Shantanu Narayen, put it in his keynote, they see their responsibility to the creative community. I think this was a wake-up call for all sides.

I would like to see other big companies care and feel a responsibility as well. Unfortunately, as I shared in just that one example above, you can see in Meta's terms that they're using our work to train AI and they don't even give you the option to opt out.

All of this is to say that Generative AI is new, it's growing, and as the latest technology, it is shaking up the industry. As a photographer, I know that there are many of us who enjoy being able to make adjustments using this tech. Many genres of creatives—even all the way up to big industries like cinema and commercial—are using AI to create. As it happens, there's a whole slew of questions and issues. But at least with Adobe, in regards to Generative AI, we can close the case on: Did they steal your photos to train Generative AI? The answer is no.

So where do you stand on how Generative AI is starting to affect the community?

Select Images by Jesus Aranguren/AP Content Services for Adobe