Adobe’s AI Terms of Service Backlash Explained

In June 2024, the creative world erupted in frustration over a software update notification. Adobe, the industry standard for design and video editing, pushed a new Terms of Use agreement to its millions of Creative Cloud subscribers. The update came with a strict ultimatum: accept the new terms or lose access to the software immediately. This triggered a massive backlash from photographers, illustrators, and film editors who feared Adobe was claiming the right to use their private work to train its artificial intelligence models.

The Spark: What the Terms Actually Said

The controversy began when users noticed specific language in the General Terms of Use, specifically regarding how Adobe interacts with user-generated content. The panic centered on sections that appeared to grant Adobe broad, invasive rights to everything a user created.

The Controversial Clauses Users flagged specific sections, notably around Section 2.2 and Section 4.2. The language stated that Adobe may “access your content through both automated and manual methods.” While the terms mentioned this was for “content moderation” and “improving our Services and Software,” the phrasing was alarmingly vague.

Furthermore, the terms required users to grant Adobe a “non-exclusive, worldwide, royalty-free sublicensable license” to use, reproduce, and publicly display content. For a freelance concept artist working on an unreleased video game, or a lawyer using Acrobat for sensitive documents, this wording sounded like a legal nightmare. It implied that Adobe could technically look at, use, or even sell access to private, proprietary work.

Why Creators Revolted

The reaction was swift and furious for three primary reasons.

1. The Fear of AI Training

The primary driver of the backlash was Generative AI. Adobe has invested heavily in Firefly, its own AI image generation model. Creators immediately interpreted the “automated methods” clause as a loophole allowing Adobe to scrape private Creative Cloud files to train Firefly. Artists feared they were paying a monthly subscription to a company that was actively building technology to replace them, using their own portfolios as the fuel.

2. The NDA Nightmare

Professional creatives often work under strict Non-Disclosure Agreements (NDAs). If a film editor is cutting a Marvel movie or a designer is working on Apple’s next product launch, they cannot allow a third party to “access” that content. The broad language in the TOS made it look like using Photoshop would violate these NDAs. High-profile users, including movie director Duncan Jones, publicly criticized the terms, suggesting that studios might have to drop Adobe products to protect their intellectual property.

3. The “Locked Out” Tactic

The delivery method worsened the situation. The update appeared as a full-screen block. Users could not uninstall the software, close the account, or even contact support without first clicking “Agree.” This forced consent tactic made users feel trapped and coerced, amplifying the distrust.

Adobe’s Response and Clarification

The outrage grew loud enough to impact Adobe’s stock sentiment and public relations. In response, Adobe’s executives, including Scott Belsky, attempted to quell the fire on social media platforms like X (formerly Twitter).

Adobe clarified that the legal language was “clunky” and had been misinterpreted. They issued a blog post and an updated FAQ to state their actual intentions explicitly.

What Adobe Claimed:

  • No AI Training on User Content: Adobe stated clearly that they do not train Firefly on user content stored locally or in the cloud. Firefly is trained only on Adobe Stock content and public domain images.
  • Clarifying “Manual Review”: They explained that “manual methods” (humans looking at files) primarily referred to scanning for illegal content, specifically Child Sexual Abuse Material (CSAM), or strictly when a user reports a bug and consents to support access.
  • Clarifying “Automated Methods”: This referred to features like generating thumbnails, virus scanning, or “content-aware fill” tools that process data to function, not to scrape data for machine learning.

The Revised Terms of Use

Following the PR disaster, Adobe promised to rewrite the Terms of Use to be more specific and human-readable. By mid-June 2024, they released updated language that explicitly restricted their own power.

Key Changes in the Revision:

  1. Explicit Prohibition: The terms now specifically state that Adobe will not use customer content to train generative AI models.
  2. Local Storage: They clarified that they do not scan files stored locally on a user’s computer.
  3. NDA Respect: The new language acknowledges that Adobe has no ownership over the user’s work and respects the confidentiality of professional projects.

The Market Consequence: Competitors Step Up

This event highlighted the fragility of Adobe’s monopoly. While Adobe managed the crisis, competitors seized the moment to market their privacy-first approach.

  • Affinity (Serif): Recently acquired by Canva, Affinity saw a surge in interest. They publicly reiterated their commitment to never charging a subscription and not using user work for AI.
  • Procreate: The popular iPad illustration app took a hard stance, releasing a statement that they would never introduce Generative AI into their app, positioning themselves as the ethical alternative for artists.

The backlash served as a warning to all tech companies. In the age of AI, vague legal terms regarding data usage are no longer acceptable. Users are hyper-aware of data scraping, and trust is becoming a defining feature for creative software.

Frequently Asked Questions

Does Adobe use my art to train Firefly? No. Adobe has explicitly stated in their updated terms and public communications that Firefly is trained on Adobe Stock images, openly licensed content, and public domain content where copyright has expired. They do not train on personal Creative Cloud files.

Can I opt out of Adobe’s content analysis? Yes. You can opt out of “Content analysis for product improvement.” This is found in your account settings under the “Privacy and Personal Data” section. However, this setting refers to machine learning for features (like organizing photos), not Generative AI training, which they claim they don’t do regardless of this setting.

Why did the terms mention a “worldwide license”? This is standard legal phrasing for cloud services. To host your file, generate a thumbnail, and let you share it with a client via a link, the software provider technically needs a “license” to display and move that data. However, the original wording lacked the necessary limitations, making it sound like an ownership claim rather than a technical necessity.

Is it safe to work on NDA projects in Photoshop? According to the revised terms from June 2024, yes. Adobe explicitly states they do not claim ownership of your content and will not access it in a way that violates confidentiality, provided you are not publishing it to a public community forum like Behance.