Adobe: we don’t train gen AI on customer content

Today in “complaints get results”: The freakout over Adobe’s recently updated Terms of Service (ToS) — which the company asked users to accept only to face a vocal backlash from some of them and vows to cancel Adobe Creative Cloud subscriptions — has led Adobe to promise a new update to its ToS by next week that it says aims to “be clearer and address the concerns raised by the community.”

Most importantly, Adobe reiterates in a blog post that: “We don’t train generative AI on customer content. We are adding this statement to our Terms of Use to reassure people that is a legal obligation on Adobe. Adobe Firefly is only trained on a dataset of licensed content with permission, such as Adobe Stock, and public domain content where copyright has expired.”

This move comes after significant public outcry over recent changes to its terms, which were initially clarified in a blog post on June 6 but failed to assuage customer concerns.

Key changes and clarifications

Adobe’s second update to the Terms of Use/Service in two weeks will emphasize several crucial points to reassure Adobe’s vast user base:

1. Content Ownership: Adobe customers retain ownership of their content. Adobe has committed that it will never use customer content to train any generative AI tools.
2. Generative AI Practices: Adobe Firefly, the company’s generative AI tool, is trained exclusively on datasets consisting of licensed content from Adobe Stock and public domain content where copyright has expired. The company reiterated that customer content is not used for this purpose.
3. Product Improvement Programs: Customers have the option to opt out of Adobe’s product improvement programs. These programs may use data and content characteristics to enhance user experiences through machine learning techniques, but not for generative AI purposes.
4. License Clarifications: Adobe will provide clearer explanations of the licenses required to operate and improve its products. These licenses will be narrowly tailored to specific activities, and Adobe emphasized that these do not transfer content ownership to the company.
5. Content Scanning: Adobe does not scan content stored locally on users’ devices. For content uploaded to Adobe’s servers, the company automatically scans for child sexual abuse material (CSAM) and other illegal content. Human reviews are conducted only if an issue is flagged or upon user request.

Response to public backlash

The backlash primarily stemmed from a clause in the Terms of Use that stated Adobe could access, view, or listen to user content through automated and manual methods to improve services using machine learning.

Many users perceived this as an invasion of privacy and a potential misuse of their content, especially for confidential or client-protected material.

In response, Adobe emphasized that these measures are in place to support innovative features like Photoshop Neural Filters and Remove Background in Adobe Express, and to ensure compliance with legal requirements.

The company clarified that it does not and cannot access content stored locally on users’ devices.

A renewed commitment to customer trust

Scott Belsky, Chief Product Officer, and Dana Rao, Chief Trust Officer, both stressed Adobe’s commitment to transparency and responsible innovation. “There is no ambiguity in our stance, our commitment to our customers, and innovating responsibly in this space,” the company stated. “We’ve never trained generative AI on customer content, taken ownership of a customer’s work, or allowed access to customer content beyond legal requirements.”

Adobe acknowledged that it should have modernized its Terms of Use sooner to better reflect the evolving technological landscape and the needs of content creators. The company has committed to continuing this dialogue with customers and incorporating their feedback into future updates.

Future initiatives

Beyond the Terms of Use, Adobe is also advancing initiatives to protect content creators. This includes efforts like Content Credentials, which provide attribution for creators and enable the addition of “do not train” tags for images shared online. Additionally, Adobe supports the FAIR legislation aimed at protecting content creators from impersonation.

Adobe recognizes that trust must be earned continuously. The company expressed gratitude for the feedback received from its community and reaffirmed its dedication to being a trusted partner for creators in the evolving digital landscape.

As its new ToS is rolled out, Adobe promised to continue to engage with its user community to ensure that their concerns are addressed and rights protected.

Clearly, Adobe hopes to rebuild trust and maintain its reputation as a responsible innovator in the digital space. Now whether customers are assuaged by this latest turn is an open question. But at the least, the company appears to be trying to respond to the objections and concerns in good faith.