On November 16, 2026, the game workers' branch of the UK trade union IWGB published a joint manifesto on generative AI use in the games industry. Besides the branch itself, members of Game Workers Unite Ireland, Spain's CSVI, and the workers' council of FIOM-CGIL Milan in Italy took part in drafting it. The manifesto covers production stages such as early concept art and first-draft translations, and asks employers to let employees take part in decisions about when and where generative AI is used. Unlike the generative AI disclosures Valve requires in Steam's pre-release review, it also addresses use during prototyping, which leaves no trace in the shipped product.
How to decide on generative AI use and refusal in the workplace
The manifesto argues that employees should decide when, where, and how generative AI is used. It proposes that employers avoid generative AI in principle and prohibit its use in early development and prototyping through a code of conduct. It also asks employers not to force employees to use it, and, where use is unavoidable, to disclose the specific purposes.
The authors are concerned with more than direct orders from managers. If deadlines are set on the assumption that work will be automated with generative AI, employees may feel pressure to use it even without an explicit instruction. They also worry about the burden of correcting AI output, and about the loss of work that junior staff once handled as it becomes automated. The manifesto says managers should be responsible for the deliverables a team completes.
The manifesto urges employees to refrain from using generative AI services where possible, and to push through their workplace representatives for negotiated agreements on the terms of use. Such agreements could include provisions protecting employees who refuse to use generative AI, and the creation of a committee to investigate violations. The manifesto's FAQ also allows for workplaces where many colleagues want to use generative AI, saying employees can decide collectively whether to allow it. The core of the proposal is to establish a procedure for setting usage policy at each workplace, along with mechanisms that protect employees who refuse to use it.
Beyond usage policy, another key issue is the use of employees' illustrations and writing to train AI. The manifesto asks employers not to use employees' work for training. Where they do, it lists as conditions the employee's explicit consent, fair compensation, and the right to have their work withdrawn from training datasets. It also asks regulators to protect these rights and to make the purposes, storage methods, and retention periods of training data transparent.
Steam's disclosure system and workplace terms for generative AI
Valve uses the Steamworks content survey, which developers submit before release, to review store pages and product builds. The survey asks about generative AI use, but it mainly covers content that players actually encounter. The manifesto, by contrast, asks for workplace discussion of the terms of generative AI use even for early materials and prototypes that do not remain in the finished product.
For example, suppose a team generates concept art with AI to explore a game's direction, uses it internally, and then discards it. That image is not part of the content players encounter. But if the same image is used in the released game, it falls under "pre-generated" content in the Steam survey. The same applies to audio and localization produced with AI during development and included in the release. For "live-generated" content, where generative AI creates content while the game is running, developers must also explain the safeguards they use to keep illegal content from being output.
What the manifesto questions, on the other hand, is whether generative AI is used at the concept-art stage at all. Whether the result is adopted in the finished game or discarded after prototyping, it stresses that employees should take part in setting policy on use during production. Steam's disclosure system is mainly a way to check the contents of products going on sale, whereas the manifesto deals with working conditions and decision-making in the production process.
The manifesto also calls on players to support games made without generative AI. For buyers, Steam's disclosures offer a clue to what AI-generated content a product contains. Employee decisions on usage policy and protection for those who refuse to use AI, meanwhile, are positioned as matters to be settled through workplace agreements, at a stage before any product review.
Growing generative AI use in games, and dependence on outside services
According to a preview of GDC's annual industry survey released in November 2026, 40% of game industry respondents use generative AI at work, up from 36% the previous year. Meanwhile, 55% said generative AI is having a negative effect on the industry, almost unchanged from 54% the year before. GDC lists research, brainstorming, early prototyping, and repetitive tasks among the uses respondents find valuable. The results suggest that while use is spreading, concern about its impact on the industry remains strong.
There is also concern that as production know-how accumulates in outside generative AI services, game developers could become dependent on particular providers. Even if adoption is cheap, once a service is deeply embedded in the production process, it may be hard to switch to another even if prices rise. The manifesto's authors likewise worry that entrusting knowledge and skills built up in-house to outside services would weaken developers' position in price negotiations with providers.
This problem arises separately from whether a prototype made with generative AI is ultimately used. Even if generated concept art is discarded, continuing to use the same service for the next prototype makes the production process itself increasingly dependent on an outside provider. Whether work can be moved to other tools when prices or terms change is also something to consider at the time of adoption.
The manifesto, including these technical and economic issues, asks that decisions on adopting generative AI not be left to management alone, but be discussed with employees in light of the effects on production processes and working conditions. As negotiations over agreements proceed at individual companies, the question will be how provisions on terms of use during prototyping, protection for employees who refuse to use AI, and the use of work for training are applied in actual production environments.
