The Electronic Frontier Foundation is urging U.S. courts to resist reshaping copyright law in response to generative AI, arguing that claims about AI-driven “market dilution” would give rightsholders excessive control over lawful competing expression.
In a new post tied to its amicus work in cases including *Concord Music Group v. Anthropic* and the *Mosaic LLM* litigation, EFF draws a line between infringement and competition. Its central argument: copyright protects specific expression, but it does not entitle a creator or publisher to stop others from making works that compete in the same market.
The legal fault line: outputs versus competition
The disputes matter because some plaintiffs contend that training and deploying generative AI systems cannot qualify as fair use if those systems enable a large volume of works that compete with copyrighted material.
EFF calls that theory a substantial departure from established copyright limits. In its view, accepting “market dilution” as a cognizable harm would let owners claim control not only over copied works, but also over non-infringing works that share a genre, trope, style, or commercial audience.
That could affect much more than foundation-model companies. Publishers, studios, creative-software vendors, independent creators and enterprise teams using generative tools could all face more uncertainty if liability turns on whether an output competes with an incumbent work rather than whether it unlawfully reproduces protected expression.
For builders, the practical distinction is significant. A system that produces a substantially similar or memorized output raises a different set of risks from one that helps users create new material in an existing category. EFF’s position is that courts should preserve that distinction rather than use copyright to police competitive pressure.
A familiar technology-policy argument
EFF frames the current debate through earlier copyright conflicts around technologies such as the VCR, player piano, gramophone and camera. The organization points in particular to the Supreme Court’s *Sony* decision, which recognized that VCRs had substantial non-infringing uses and warned against judicially remaking copyright rules to address a new technology.
The analogy is not a claim that AI’s effects will mirror those of home video. Generative systems raise distinct questions about training data, output similarity, attribution and compensation. But EFF’s message to courts is institutional: broad, general-purpose tools can have lawful and unlawful uses, and speculative predictions about market disruption should not by themselves expand exclusive rights.
What operators should watch
The immediate issue is not whether AI companies are free of copyright exposure. It is which legal theories courts validate as these cases move forward.
Operators should watch for rulings that clarify:
- whether and when model training is treated as fair use;
- how courts evaluate evidence of memorization or substantially similar outputs;
- whether alleged downstream competition counts as copyright market harm absent infringement; and
- how courts distinguish protected expression from unprotected styles, ideas, genres and tropes.
Those outcomes will shape product design and procurement. Model providers will continue to need safeguards around output similarity, user reporting and rights-management workflows. Enterprises adopting AI should retain review processes for externally published creative work, especially where prompts, reference files or outputs involve recognizable third-party material.
At the same time, a ruling that embraces broad market-dilution claims could raise the cost and legal uncertainty of building or deploying creative AI products—even where outputs are not infringing. EFF’s intervention is therefore a warning against treating copyright as a general mechanism for insulating established creative businesses from new forms of competition.
The larger policy debate over creator compensation and AI training data is unlikely to disappear. EFF’s argument is that courts should address it within copyright’s existing boundaries, rather than create a new right to block lawful expression because it may compete.
