Upon discovery of Adobe’s updated terms of service in February, users raised their voices in protest. Users were informed by Adobe that it could access their content “through automated and manual methods” and utilize “approaches like machine learning to enhance [Adobe’s] Services and Software.” Many interpreted the update as compelling users to provide unrestricted access to their work for the purpose of training Adobe’s generative AI: Firefly.
In a clarification issued by Adobe on Tuesday, the updated terms of service agreement pledged not to train AI on user content stored locally or in the cloud and provided users with the choice to opt-out of content analytics.
Involved in the crossfire of intellectual property lawsuits, the previous ambiguous terms update shed light on a prevalent atmosphere of doubt among artists, many of whom heavily rely on Adobe for their artistic endeavors. Jon Lam, a senior storyboard artist at Riot Games, remarked, “They have already shattered our confidence,” in response to the revelation that award-winning artist Brian Kesinger detected generated images resembling his art style being marketed under his name on Adobe’s stock image platform without his authorization. Recently, the estate of the deceased photographer Ansel Adams publicly admonished Adobe for allegedly selling generative AI replicas of his work.
Efforts by Scott Belsky, Adobe’s Chief Strategy Officer, to alleviate concerns among protesting artists were met with skepticism. He clarified that machine learning pertains to the company’s non-generative AI tools, such as Photoshop’s “Content Aware Fill” tool, which enables users to seamlessly erase objects from an image. Despite Adobe’s claims that the revised terms do not confer content ownership to the company and that user content will not be utilized to train Firefly, the misinterpretation sparked a broader discourse on the company’s dominant market position and how such modifications could jeopardize artists’ livelihoods at any given moment. Lam is one of the artists who still harbors doubt and believes that, despite Adobe’s reassurances, the company may exploit works generated on its platform to train Firefly without the creators’ consent.
Apprehensions regarding the unauthorized utilization and commercialization of copyrighted content by generative AI models are not novel. Last year, artist Karla Ortiz encountered images of her work being produced under her name by various generative AI platforms, leading to a class-action lawsuit against Midjourney, DeviantArt, and Stability AI. Ortiz’s experience was not unique—Polish fantasy artist Greg Rutkowski discovered that his name was frequently used as a prompt in Stable Diffusion upon its initial launch in 2022.
Having dominated the industry for over three decades as the creator of Photoshop and PDFs, Adobe has been the go-to standard for the majority of creatives. An attempt to acquire the product design company Figma was halted and abandoned in 2023 due to antitrust concerns, underscoring its sheer scale.
Adobe insists that Firefly is “ethically trained” on Adobe Stock, yet Eric Urquhart, a long-standing stock image contributor, disputes this claim, asserting that “there was nothing ethical about how Adobe trained the AI for Firefly,” highlighting that Adobe lacks rights to any images from individual contributors. Urquhart originally shared his images on Fotolia, a stock image platform, where he consented to licensing terms that did not mention any provisions for generative AI usage. Adobe later acquired Fotolia in 2015 and implemented undisclosed terms of service updates, enabling the company to train Firefly using Eric’s photos without explicit consent: “the wording in the current alteration of TOS closely resembles what I observed in the Adobe Stock TOS.”
