Adobe Sold Firefly as the Ethical Choice, Then Got Caught Training It on Competitors' Images

What happened

Adobe built its pitch for Firefly on a specific promise: unlike its rivals, this AI image generator was trained on clean data. The company drew a clear line between itself and competitors such as DALL-E 3, Stable Diffusion, and Midjourney, all of which faced criticism and litigation for training on artist images without consent. Firefly's training set, Adobe said, came primarily from licensed Adobe Stock images and public domain material. The company even created a bonus compensation scheme for artists whose work contributed to the first release, a gesture designed to signal that this was a different kind of product from a different kind of company.

In April 2024, a Bloomberg report undercut that premise. According to the report, roughly 5 percent of the images submitted by those compensated artists came from competitor AI image generation systems, not from original human-made photographs or illustrations. Whether those competitor outputs were themselves generated using copyrighted material was, at the time of the report, unclear. That ambiguity matters, because the competitors whose tools contributed to Firefly's training data are the same tools Adobe had positioned itself against for their copyright problems.

The chain of concern is short and direct. If Midjourney or Stable Diffusion trained on copyrighted images without permission, and artists submitted Midjourney or Stable Diffusion outputs to Adobe Stock, and Adobe then trained Firefly on those submissions, then the commercial safety Adobe was advertising rested partly on a foundation it had not verified. The bonus compensation scheme paid artists for contributing images, but it did not verify what those images actually were.

Adobe's response, as reflected in the coverage, did not include a specific denial of the reported figures or an alternative account of how contaminated submissions were handled. The episode drew descriptions of "ethics-washing," a term applied when an organization uses ethical language and positioning to create a market advantage it cannot fully substantiate. Observers noted that the marketing characterized Firefly as a safe choice for commercial use without the mechanisms in place to ensure that claim could survive scrutiny.

What the incident surfaces is a gap in how training data provenance is documented and confirmed. A company can make accurate-sounding claims about what went into a model, offer a compensation program as evidence of good faith, and still be unable to produce a provable record of what a system was actually trained on. Without that kind of verifiable trail, "ethically trained" functions as a differentiating label rather than a checkable fact, and the label holds only as long as no one looks closely at the underlying data.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.