Does Adobe train AI on your work? What its terms say in 2026

The Morse teamPublished October 26, 20269 min read

No, not unless you submit it to Adobe Stock. As of October 2026, Adobe's General Terms of Use say it will not use your files, on your computer or in its cloud, to train generative AI models such as Firefly, except content you choose to submit to the Adobe Stock marketplace. Separately, Adobe may still run "content analysis" on files stored on its servers to improve non-generative features. Adobe says minors and business accounts are "automatically opted out" and doesn't state the default for adult personal accounts, so we'd assume yours is on until you turn it off at account.adobe.com/privacy.

We read Adobe's General Terms of Use, its content analysis FAQ, the Firefly FAQ, the Adobe Stock contributors' Firefly FAQ, its two June 2024 blog posts on the terms, and its Content Credentials help pages on October 26, 2026. Every fact below links to the Adobe page it came from. Adobe deserves more credit here than it usually gets. It wrote the no-training promise into its binding terms, pays the Stock contributors whose work it trains on, and opts minors and business accounts out of content analysis automatically.

Does Adobe use your files to train AI?

Only files you submit to Adobe Stock. Adobe's General Terms of Use, effective October 3, 2025, say: "We will not use your Local or Cloud Content to train generative AI models except for Content you choose to submit to the Adobe Stock marketplace." Every other setting Adobe offers covers something narrower.

Adobe's settings, read October 26, 2026:

The same terms add one more exception in section 4.3: Adobe won't train generative AI on your content "except at your specific request (like you asking us to train a custom model on your Content)." This isn't legal advice, so read the terms yourself before you sign a client contract that depends on them. For how Adobe compares with chat apps, see which AI companies train on your data.

What is Adobe content analysis?

It's machine learning on your cloud files to improve Adobe's own features, and it doesn't train generative AI. The content analysis FAQ, updated June 4, 2026, says "We do not analyze your content to train generative AI models, unless you choose to submit content to the Adobe Stock marketplace." Adobe's June 10, 2024 post gives masking and background removal as the kind of features it builds this way, "through techniques including machine learning (NOT generative AI)."

The terms call this Content Analytics: analysis of how you use Adobe's apps and "the characteristics of your Content (such as file type and structural attributes)." Adobe may hand that work to "trusted vendors and contractors subject to confidentiality and privacy restrictions." Section 2.2 also says "Insights from Content Analytics may be used to inform our marketing to you."

Several of the top results still date from January 2023, when How-To Geek ran the headline "Adobe Is Using Your Data to Train AI." Adobe says it has "never trained generative AI on customer content" (Adobe), and its FAQ excludes generative training whether the toggle is on or off. Whether your work trains Firefly comes down to whether you upload it to Stock. Turning off content analysis is still worth doing.

How do you turn off Adobe content analysis?

Sign in at account.adobe.com/privacy, find the Content analysis for product improvement section, and click the Off toggle (Adobe). On the same Data and privacy page, you can also turn off Desktop app usage, which covers how you use features rather than what's in your files (Adobe).

You won't see the toggle on a team, company, or school account, because Adobe says those are "automatically opted out." A minor who doesn't see it is opted out too (Adobe). For an adult's personal account, the FAQ says only that you "can turn off content analysis for product improvement at any time" and that, "subject to your opt-out and consent rights," Adobe "may analyze your content that is processed or stored on our servers." It never says the setting starts off, so we read it as on until you change it. Check the toggle rather than take our reading on trust.

The opt-out has edges. Adobe's FAQ says that "when you are not logged in" or when you join a program where you submit content, turning it off "does not affect our ability to analyze your content for product improvement." It names four such programs: the Adobe Photoshop Improvement Program, feedback you send with content, betas and user studies, and Adobe Stock. So sign in before you work, and leave beta features and feedback forms alone for client files.

Is Firefly trained on user content?

Not on Creative Cloud subscribers' personal content, Adobe says, but it is trained on Adobe Stock. The Firefly FAQ, updated May 13, 2026, answers "is my content automatically used to train Firefly?" with "No, we don't train on any Creative Cloud subscribers' personal content. For Adobe Stock contributors, the content is part of Firefly's training dataset." It describes the data as "Adobe Stock images, openly licensed content, and public domain content where copyright has expired." Adobe's approach page adds: "We do not mine the web or video hosting sites for content."

"Licensed content" covers more than photographs. The Stock contributors' Firefly FAQ, updated September 16, 2026, bases the 2026 Firefly bonus, paid September 17, 2026, on "photos, videos, vectors, illustrations, and Generative AI content considered for training between June 3, 2025, and June 2, 2026." So AI-generated images that contributors submitted to Stock are in the pool. The same page says "Adobe Stock video is now used in training," and "We currently do not offer an opt-out option for Adobe Stock content."

Two Firefly features take your images or prompts without feeding them to Adobe's models. Adobe's custom models page says it "only uses the images you upload to train your custom model" and that they're "never used to train Firefly's foundational models." Training one costs 500 credits, and the model is private by default. The partner models page says that for models like GPT Image and Nano Banana inside Firefly, "Adobe and our partners do not train on your personal or project data." We compare Firefly with other image apps in our private Midjourney alternatives.

What changed in Adobe's terms of service in 2024?

The wording changed, and Adobe says the practice didn't. Here are the dates from Adobe's own pages:

• June 6, 2024: after a routine re-acceptance prompt drew questions from users, Adobe's clarification said the update reflected that it had "added more human moderation to our content submissions review processes."
• June 10, 2024: Adobe promised new terms by June 18, saying "We've never trained generative AI on customer content" and that it would add that statement "to reassure people that is a legal obligation on Adobe."
• October 3, 2025: the current General Terms of Use took effect, with the no-training line in sections 2.2 and 4.3.
• June 4, 2026: Adobe updated the content analysis FAQ, saying it clarified "how your opt-out rights work."
• September 17, 2026: Stock contributors whose work was considered for Firefly training got this year's bonus (Adobe).

The terms can change again. Section 1.5 says Adobe may update them "from time to time" and will notify you of important changes (Adobe).

Can people at Adobe see your files?

Sometimes, for files on its servers. The terms say Adobe doesn't "scan or review content that is stored locally on your device," but cloud files "may be automatically scanned" for illegal content such as child sexual abuse material, and a person may look if a file is flagged. Human review may also happen when you ask for support, publish on Stock or Behance, or join a beta or improvement program. The content analysis FAQ adds work submitted to be featured on Adobe Express and Lightroom tutorials.

Prompts are in scope too. Adobe's Generative AI User Guidelines, updated May 15, 2026, say your prompts, inputs, and results "may be reviewed through both automated (such as machine learning) and manual methods for abuse prevention and content filtering purposes."

Does a "do not train" Content Credential stop AI training?

It's a request, and Firefly honors it. Adobe's preference page, updated September 2, 2025, lets you embed "I request that generative AI models not train on or use my content" in a file's Content Credentials, and says "Currently, this preference is supported by Adobe Firefly." You need a verified name or a connected social account to set it. A file carrying it can't be uploaded to Adobe Stock or used in Style Reference, Structure Reference, or Generate Similar. Other companies' models aren't bound by it.

How Morse keeps your creative work out of training

Every exception on Adobe's list starts with your work sitting on Adobe's servers: Stock uploads, cloud files open to analysis, betas, feedback, and anything you do signed out. Morse is built so your work doesn't sit on ours. Morse doesn't train AI on your content, sell your data, or show ads, and your chat history and the images you make in Studio, Morse's image generator, are stored in your browser, encrypted (privacy policy).

There's no Stock-style exception, because Morse doesn't train AI on anything you write or make. There's no content analysis toggle to find, because Morse's servers don't keep your messages or log what's in them. Files you attach are sent with the message and never stored by Morse.

Each chat has its own privacy mode:

• Private, for drafting briefs, captions, and concepts no host should keep. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run on hosts that agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• E2EE, on Pro, for unreleased work and client notes no one should read. Your browser checks the enclave's proof that it's genuine before anything leaves your device, then encrypts the message so only that enclave can read it. Morse's servers relay text they can't read. It rests on the enclave hardware and attestation working as designed, and it runs open-weight models in chat, without image generation or web search.
• Anonymous, for GPT, Claude, and Gemini without the provider knowing it's you. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the message and may keep it for a limited time to check for abuse, under its own policy, so leave client names out.

In every mode, history stays in your browser, locked when you sign out and opened with two keys, one of which never leaves your device. Clearing your browser deletes it for good and Morse can't recover it, so keep the originals of your files elsewhere. Because we don't store chats, we usually have very little to give authorities.

What to do, by concern

• Firefly training: don't upload anything to Adobe Stock that you don't want in Firefly's training set, since Stock has no opt-out. Morse doesn't train AI on what you write or make.
• Content analysis: turn off Content analysis for product improvement at account.adobe.com/privacy, and stay signed in while you work.
• Other companies' models: add the Content Credentials "do not train" request to work you post, knowing it's a request.
• Client work in an AI chat: use an E2EE chat on Morse, so our servers relay only text they can't read. Writers can read more on where a manuscript goes when you paste it.

Try a chat from the Morse home page without an account, or start free with 3 Studio images a day. End-to-end encrypted chats come with Pro, at $19 a month.