Creative Workers Want More Than “Consent”: Governance of Generative AI in Creative Work

This paper does not treat the conflict between generative AI and creative labor simply as a copyright infringement issue. The authors interview 20 creative workers from visual art and design, writing, and programming to ask what kind of AI governance creative workers actually want. The central argument is that the principles of “consent, credit, and compensation,” often called the 3 Cs, are necessary but not sufficient.

The 3 Cs Require Conditions, Not Just Slogans

In debates around generative AI, three demands appear repeatedly. If my work is used for training, ask for my consent. If my work has been used, give me credit. If someone profits from my work, compensate me. At first glance, these seem like straightforward principles of fairness. The most interesting part of this paper begins precisely here. The authors do not reject the 3 Cs. Rather, they show that once these principles are placed inside the actual conditions of creative labor, they become far less stable than they first appear.

The first principle to become unstable is consent. Most interview participants said they wanted to be asked before their work was used to train AI models. But if consent becomes a one-time legal checkbox, it creates new problems. Today’s AI and the AI of five years from now may not be the same tool. A use that seems acceptable today because it merely generates examples or drafts may later become the foundation of a model that replaces the worker’s own job. For this reason, the authors treat consent not as a fixed contract clause, but as a process that must be renewed as the technology changes.

The more important issue is power. Creative workers employed by companies are often not in a position to say “no” to the use of their work. If the company legally owns the output, the worker is treated less as a subject of consent and more as someone who has already handed over their rights. Freelancers are not free from this problem either. In an environment where work opportunities decline and rates fall, refusing to allow one’s work to be used for training may lead to real economic disadvantage. Consent, then, cannot be understood only as an individual choice. To be meaningful, it must include the ability to refuse without punishment, clear explanation, time for consideration, the possibility of withdrawal, and collective channels for negotiation.

Credit Is Not Always a Good Form of Reward

The discussion of credit is even more subtle. In ordinary creative ecosystems, being named matters. Attribution is tied not only to recognition but also to career development, reputation, and future work. In the context of AI training, however, attribution does not always benefit the creator.

Some participants said they were not famous enough for attribution to matter. Others thought that if the work was created inside a company, the company rather than the individual should receive the credit. More importantly, some participants worried about responsibility. Once their work is absorbed into training data, they cannot control what kinds of images, sentences, code, political propaganda, hate speech, or low-quality outputs may be generated from it. If their name becomes attached to such a model or output, credit turns into risk rather than honor.

This is where the paper carefully distinguishes among attribution, misattribution, non-attribution, anonymity, and the integrity of the work. What creators want is not simply to have their names attached. They want the right to decide how their work is connected to AI systems. If having one’s name attached to AI-generated output results in distortion or reputational harm, credit becomes another form of violation rather than a protective measure.

Compensation Is About Money, But Not Only Money

Participants were relatively clear about compensation. If their work is used to train AI models, they want to be compensated. But here too, there are distinctions. Some company employees felt that work created during paid working hours had already been compensated through salary. Yet they also believed that if the same work continued to be used for training or derivative outputs after they left the company, additional compensation would be necessary.

For freelancers, the issue is more direct. If a platform or publisher sells a creator’s work, then uses that work to train AI, and later profits from derivative outputs, the existing payment structure no longer explains the situation adequately. A single payment for a commissioned work cannot automatically be assumed to include unlimited future reuse.

Another important distinction is between professional work and personal projects. Participants tended to feel less attached to work produced for companies, while reacting more strongly to the use of personal projects. This reveals that creative work is not merely data. Some work is produced for a livelihood, while other work carries traces of self-formation. Even when compensation is offered, some uses may remain unacceptable to the creator. Money is necessary, but it does not erase every feeling of violation or discomfort.

AI Governance Does Not Belong Only to Governments

One of the strengths of the paper is that it does not reduce governance to state regulation. The authors also treat companies, publishers, and freelance platforms as important governance actors. In practice, the rules creative workers encounter often appear first not as law, but as contracts, platform policies, company guidelines, submission rules, and internal meetings.

However, the interviews show that many organizations still lack clear policies on generative AI. Some respond only after problems arise. Others simply ban AI use altogether. The better governance imagined by the authors is neither simple prohibition nor blanket permission. It requires clarity about when AI may be used, what kinds of work should not involve AI, how outputs should be labeled, whether work will be used for training, how long it will be stored, and what happens to that work after the worker leaves the organization.

Platforms are especially important here. Some participants pointed out that AI-generated content is increasingly entering image libraries and freelance platforms while pretending to be human-made work. This is not merely a matter of taste. It affects how designers search for references, how clients choose workers, and how creators sell their services. This is why labels, filters, badges, and verification systems that distinguish human-made work from AI-generated work matter.

The Ethical Boundary Lies Between Replacement and Augmentation

Participants do not reject generative AI altogether. Many see AI as useful for reducing repetitive tasks, producing drafts, lowering technical barriers, and helping people learn complex fields. AI can assist with maintaining consistent linework across animation frames, or help programmers understand difficult codebases.

The problem lies in whether AI is used to expand the judgment and skill of creative workers, or to replace their labor with cheap outputs. The authors treat this distinction as crucial. There is a major difference between using AI to handle repetitive labor in the background while humans retain direction, standards, context, and responsibility, and using unauthorized training on human creative work to push those same workers out of the market.

This difference leads to a broader discussion of creative quality. Participants argue that AI can quickly produce outputs, but the ability to define problems, read context, and take responsibility for meaning still depends on human training and experience. Design is not simply the production of attractive graphics; it is the solving of problems. Writing is not merely the arrangement of sentences; it is the construction of perspective and judgment. Programming is not the act of pasting code; it is the understanding of systems. The more AI produces easy outputs at scale, the more invisible the time and training behind creative work may become.

The Future of Creative Labor Is Already Being Decided in the Data Pipeline

One of the paper’s important insights is that the risks of AI should not be examined only at the output stage. AI regulation often focuses on where AI is applied. Recruitment, evaluation, surveillance, medicine, and finance are often classified as high-risk areas because their outcomes affect people directly. But the authors argue that the process through which AI is built—the collection of data and the training pipeline—also affects labor markets.

If creative workers cannot know how their work is collected, under what conditions it is stored, which models it enters, and what commercial services it later supports, they cannot understand where and how their labor is operating. This is where information asymmetry emerges. Large technology companies possess data, models, contracts, and infrastructure. Creative workers may not even know whether their work has already been absorbed. For this reason, the authors argue that AI governance must address not only outputs, but also training data, data pipelines, and organizational conditions of use.

This perspective changes the question in generative AI debates. The important question is not only “Can AI create like a human?” A more pressing question is “Under what conditions does AI absorb and reorganize human creative labor?” Once the question shifts in this way, the center of the debate also moves from technical capability to rights, contracts, platforms, labor conditions, and collective bargaining.

From Individual Rights to Collective Negotiation

The authors ultimately move toward a dual position: individual creative workers need stronger rights, but individual choice alone is not enough. It is difficult for a single creator to negotiate consent, credit, and compensation with a large platform or technology company. Freelancers in particular often struggle to form unions or engage in collective bargaining. This is why the paper points toward the need for intermediate structures such as workplace councils, professional associations, standard contracts, and royalty frameworks.

At this point, the paper does not portray creative workers simply as vulnerable subjects in need of protection. Rather, it argues that they must participate in AI governance. The perspective of creative workers is not merely anti-technology sentiment. It is field knowledge about how creative work is made, circulated, evaluated, and monetized. Without this knowledge, AI policy easily becomes an abstract ethical declaration or a belated response to damage already done.

The Core Meaning of the Paper

The value of this paper lies in its refusal to frame the conflict between generative AI and creative labor as a simple opposition between being for or against AI. The authors do not claim that creative workers are merely afraid of AI. Nor do they optimistically treat AI as nothing more than a useful tool. Creative workers can use AI. But when their work is trained on without their knowledge, when their names are attached to outputs they did not approve, and when their labor becomes a source of future profit without any decision-making power, the problem is no longer tool use. It is the distribution of power.

The 3 Cs remain important. But what this paper shows are the concrete conditions under which they matter. Consent must be renewed, and refusal must not result in punishment. Credit must include not only the right to be named, but also the right not to be named and the right not to be distorted. Compensation must account not only for the initial delivery of work, but also for future use, derivative profits, and post-employment exploitation. None of this can be left entirely to individual goodwill or corporate self-regulation.

In the end, this paper makes visible what disappears the moment creative work is called “data.” Work contains time, skill, reputation, attachment, livelihood, and responsibility. If AI governance cannot address these elements, creative workers will remain raw material suppliers rather than beneficiaries of technological progress. The kind of governance this paper calls for is closer to a set of conditions for moving creative work fairly into the age of AI. It is not an argument for stopping technology. It is an argument for following technology’s movement closely enough to see who loses what, who gains what, and who gets to decide.