Why game developers distrust generative AI and what the data says

Rick Vu, 9 October 2026

An AI agent drafted this piece under Rick's direction, as part of an experiment in AI writing. Every figure links to its source.

Game developers use generative AI and distrust it at the same time. In GDC’s 2026 State of the Game Industry report, 36% of more than 2,300 industry professionals said they use generative AI tools at work, and 52% said the technology is having a negative impact on the industry. Most of the use is research, email, and code help, while the distrust is about training data, jobs, and generated content in shipped games.

This article puts the 2026 figures next to each other and then sets out the reasons developers give. One disclosure comes first. Rick Vu, who directs this section, builds Red Dog Studio, a game studio building AI-native NPCs, so he has a stake in how the debate turns out.

The 2026 numbers side by side

GDC has asked the same question for three years, and opinion has turned fast. The share of respondents who see a negative impact rose from 18% two years earlier to 30% in 2025 and 52% in 2026, according to GDC’s 2026 State of the Game Industry report. The share who see a positive impact fell from 13% in 2025 to 7%.

Use did not fall with opinion. The same GDC report puts use at 36% overall. At game studios it is 30%, and at publishers, support companies, and marketing and PR firms it is 58%.

The people who make the content hold the most negative views. In GDC’s report, 64% of visual and technical artists, 63% of game design and narrative staff, and 59% of programmers see a negative impact. Executives were the most positive group, at 19%.

Steam gives a second view, based on what developers ship. In June 2026, Dataconomy’s report on Steam Next Fest demos found that 1,704 of the 8,700 demos in the latest Next Fest carried a generative AI disclosure, which is 19.5%. The count came from SteamDB, by way of Eurogamer.

Players are the hardest group to measure. The figure most often quoted comes from Inworld AI’s Gamer Attitudes to NPCs report, as covered by PocketGamer.biz. It found that 99% of 1,002 US gamers aged 16 to 50 expected advanced AI in NPCs to have a positive impact on gameplay, and 81% were open to paying more for such a game. It is a vendor-commissioned survey. Inworld sells AI NPC technology and released the findings itself in February 2023. The survey also asks about smarter characters, not about generated art. Read it as the most favorable version of the question, from the company with the most to gain.

Why use and distrust both run high

The GDC report does not say how many respondents both use generative AI and think it does harm. So you cannot subtract one figure from the other. Two readings fit the data, and they can both be true.

The first reading is that different people answer each question. Use is highest outside studios, at 58%. Negative views are highest among artists, designers, and programmers. Executives are the most positive group. Part of the gap is between the people who buy and manage the tools and the people whose work the tools imitate.

The second reading is that the same people use it for different tasks. Among users in GDC’s report, 81% use it for research or brainstorming, 47% for daily tasks such as email, 47% for code help, and 35% for prototyping. None of that reaches a player. A developer can draft an email with a chatbot and still object to generated textures in a released game.

One reading is that the second explains more of the gap. Developer distrust is mostly about generated content that players see and about how the models were trained. It is much less about a chatbot that answers a question.

Steam’s rules draw the same line. Valve does not count tools that only make development faster, so the 19.5% figure counts content players see. It says nothing about how many of the other demos used a code assistant. The figure is also self-reported, so it measures disclosure, not use.

What Valve now asks developers to disclose

The Steamworks content survey documentation splits AI content into two kinds. Pre-Generated AI content is made with AI tools during development and ships with the game. Live-Generated AI content is made while the game runs. For live content, Valve also asks developers to describe the guardrails that keep it from producing illegal content. On tools, the page says that “Efficiency gains through the use of these tools is not the focus of this section.”

In January 2026, Valve rewrote the disclosure form. Game Developer’s report on the Steam AI disclosure change says the form now has two parts. In the first, developers describe in a text box any AI used to make content for the game, its store page, or its marketing. The second is a checkbox for games that generate images, text, or other content during play. AI coding assistants and similar workflow tools stay exempt.

The reasons developers give

GDC’s public summary gives figures but no reasons. Press coverage of the survey, including TechSpot’s, reports respondent comments, and this article could not retrieve every quoted passage. The reasons below are paraphrased from that coverage and from the survey’s layoff data.

Training data and consent

Some respondents describe generative AI as theft, and coverage such as Time Extension’s leads with that framing. The argument is that image and text models learned from work that artists, writers, and programmers published, without their consent and without payment. For people who hold this view, a better model does not fix the problem, because the objection is to how the model was made. That fits the finding that artists, whose portfolios are part of the training data, are the most negative group.

Jobs and layoffs

GDC’s report found that 28% of respondents had been laid off in the past two years, and 33% in the United States. Half said their current or most recent employer had made layoffs in the past 12 months. The survey does not show that generative AI caused those layoffs. Still, developers watch teams shrink while executives, the most positive group in the data, promote the tools.

Quality

Developers also object to the output itself. A model produces the likely output for its training data, so generated art and writing drift toward the familiar. Games sell on a distinct look and voice. A person also has to check generated output, and that review time is often missing from estimates of the savings.

Player backlash

Coverage of the survey, including TechSpot’s article, says the outcry is loudest when generative AI makes assets in a game, and names Black Ops 7, Anno 117, and Arc Raiders. For a small studio, this turns an ethical question into a commercial one. As reasoning, not data, a disclosure may cost some wishlists, and a hidden use that players find later costs more.

The other side

Not every respondent agrees. Some respondents, per Hitmarker’s coverage, call the controversy a temporary “moral panic,” and 7% of respondents see a positive impact. The data suggests that most developers object to particular uses. Their objections focus on generated content in shipped games and on training data used without consent. Many of the same people accept AI for research and code.

How a small studio should disclose AI use

Every page in this section carries a note that an AI agent drafted it under Rick Vu’s direction. That note is a small working example of a disclosure, and it has the parts a studio needs.

The note uses a specific verb. “Drafted” tells the reader what the tool did, and “made with AI” does not. It also names the person responsible, so the reader knows who made the choices. And it sits on the page itself, where the reader decides how far to trust the text, instead of on a policy page.

A game disclosure needs the same parts, plus the detail Valve asks for:

1. List every place where generated content ships, such as art, audio, voice, text, localization, store page images, and trailers.
2. For each one, say what the tool made and what a person did after. For example, “concept art generated, then repainted by our artist” tells players more than a checkbox does.
3. If the game generates content while it runs, say what it generates and what limits apply to it.
4. Put the same text on the store page, in the credits, and in the press kit.
5. Update the text when your use of the tools changes.

Steam does not require you to list code assistants. You can mention them in one line if players ask. Do not use that line to hide generated art among the tools.

The note on this page reads as follows. “An AI agent drafted this piece under Rick’s direction, as part of an experiment in AI writing. Every figure links to its source.” The two sentences together give a reader enough to judge the article.

Related reading

• How AI NPCs work, for what live-generated content means inside a game.
• What is an AI-native game studio, for studios whose games depend on content generated at runtime.
• Game design lifecycle vs software lifecycle in the age of AI coding, Rick’s own essay on where AI tools fit in the work.