Adobe to update terms of service despite backlash

Adobe is updating its terms of service after ambiguous language sparked backlash from users regarding privacy and ownership of their work.

In a blog post On Monday, the company that makes Photoshop, Premier and InDesign creative software tools said it would roll out updated language to its terms of service by June 18, 2024. “At Adobe, there is no “no ambiguity in our position, our commitment to our customers, and to innovate responsibly in this area,” wrote executive vice presidents Scott Belsky, who oversees product, and Dana Rao, who oversees legal and policy aspects.

The tweet may have been deleted

Belsky and Rao wrote that their company had “never trained generative AI on customer content”, nor taken ownership of people’s unpublished works, and that Adobe had not announced plans to do these things with the recent ToS update. It should be noted that Adobe’s Firefly generative AI models are trained on contributions to Adobe’s archive library as well as public domain data, but this is processed separately from user-created content at their own personal and professional purposes.

“That said, we agree that evolving our Terms of Service to reflect our commitments to our community is the right thing to do,” Belsky and Rao wrote.

Last week, a PR fiasco erupted when users were informed of Adobe’s updated terms of service. Without a clear explanation of what the changes meant or what was changed, Adobe users assumed the worst and felt that the updated terms meant drastic changes to the autonomy of their content. Specifically, users believed that Adobe could now access previously unpublished work to train its Firefly AI models and could even assume ownership of work in progress. The updates lacked clarity and transparency at a time when generative AI tools are seen as threatening the work and livelihoods of creatives. Backlash followed, including promises to abandon the platform.

But it turns out that Adobe’s updated policy that grants the company access to user content was intended to detect activity that violates the law or violates its terms. Adobe said it never intended to train its models on user content or usurp any control. Belsky and Rao also claimed that users have the choice to “opt out of its product improvement program” (sharing content for training models), that its licenses are “narrowly tailored to necessary activities” such as looking for inappropriate or illegal behavior, and that Adobe does not analyze users’ content stored locally on their computers.

So, all of this could have been avoided with clearer communication, but the company’s reputation was likely tarnished.

“We recognize that trust must be earned,” Belsky and Rao said in closing the message.