
The process wall
In early June 2024, Adobe pushed a routine re-acceptance of its Terms of Use to Creative Cloud and Document Cloud customers. The prompt itself, not a product change, was the trigger: creators reading the re-acceptance screen for the first time found broad licensing language about accessing their content and read it as permission for Adobe to train generative AI on their files. Adobe's first public response, posted 6 June 2024, frames the update's actual purpose as 'improvements to our moderation processes' for content submissions, and states that 'Adobe requires a limited license to access content solely for the purpose of operating or improving the services and software and to enforce our terms and comply with law.'
What the documents show
That first statement did not settle the reaction. A follow-up post dated 10 June 2024 goes further, stating plainly that Adobe has 'never trained generative AI on customer content, taken ownership of a customer's work, or allowed access to customer content beyond legal requirements,' and commits to publishing revised, plainer-language terms 'by June 18, 2024.' It also points to Content Credentials as a parallel mechanism, describing it as enabling 'the addition of do-not-train tags for images shared online.' The current Adobe General Terms of Use, as retrieved on 16 September 2026, carry the same commitment forward in contract language: Adobe '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,' and states 'Your Local Content is never reviewed by us.'
Authorship and limitation
The episode is a useful case for a distinction makers routinely blur: a licence a company grants itself to operate a service, render a file, scan for abuse, back it up, is not a licence to train a model, and a company's promise not to train is not itself a change in copyright law over the work. Adobe's terms are a contract Adobe can amend again; they bind Adobe's conduct, not the copyright status of what a user uploads, and Adobe's posts are Adobe's account of its intentions, not an independently audited practice. The carve-out for Adobe Stock submissions in the current terms marks the one case where a user's opt-in, submitting to a marketplace, does open a training pathway.
What survives
Editorially, a Creative Cloud user keeps the ability to read a re-acceptance screen literally rather than by reputation, checking the current terms' Stock carve-out before assuming 'not trained on' covers every use of an account. Content Credentials tagging is a separate, optional mechanism attached to individual files, not something the Terms of Use grant by default.
- Read the current terms' Stock-submission carve-out against your own workflow.
- Compare the 6 June and 10 June statements to see how the company's own explanation changed under pressure.
- Check whether Content Credentials' do-not-train tag is attached to files before sharing them publicly.
Adobe's own record shows a company narrowing and clarifying a contract in public, in under two weeks, without altering the separate question of what copyright law already permits or forbids.
Sources & reading trail
States the update's purpose was moderation-process changes and describes Adobe's limited licence to access content.
Source published: 6 June 2024 · Retrieved: 16 September 2026
States Adobe has never trained generative AI on customer content and commits to revised terms by 18 June 2024.
Source published: 10 June 2024 · Retrieved: 16 September 2026
Current contract language on the generative-AI training restriction and the Adobe Stock carve-out, as a living document.
Source published: Not established · Retrieved: 16 September 2026
Papers, documentation, rulings and the makers' own records establish the note; the authorship reading and what-survives are Creative AI editorial analysis. This retrospective draft does not imply the site published on the event date.