House House discloses minimal AI use in Big Walk as IGF jurors reject any generative AI
A bug fix worth 10 lines of code has split judges of the Independent Games Festival over whether any use of generative AI should rule a game out.
House House, the developer of the co-operative puzzle exploration game Big Walk, has publicly disclosed that it used large language models near the end of development. The studio says the use was limited to "well under 0.1%" of the game's code. The disclosure came after 19 members of the Independent Games Festival jury, which has more than 50 members, released a signed statement saying they would decline "to vote for any game that uses generative AI in any capacity."
The case matters because the amount of AI involved is so small. It tests whether the line for legitimate creative work runs at how much AI was used, or at whether it was used at all. Ars Technica reports that jurors had been discussing the issue in private groups for months before it became public this week.
The jury statement
The jurors' statement did not name any games. It cited "a significant increase in the number of submitted games developed with generative AI tools" and listed ecological, economic, psychological and labour harms. The signatories said the use of these tools is still "not a foregone conclusion in our industry." They also asked the IGF to "adopt a strong anti-AI policy for future years that disallows games containing any AI-generated work."
By Tuesday, Big Walk was widely discussed as a central case. House House gave a statement to the outlet Aftermath and then added it to the game's FAQ page. The studio said "LLMs were used sparingly" to "help work through some difficult technical challenges and bugs."
What the studio says it did
According to House House, the main problem was a bug in the Unity engine that slowed the game down by producing shadows of "infinite length" in some situations. The studio says it consulted a Unity engineer, who concluded a fix "wasn't possible." It then asked an LLM chatbot, which suggested a possible fix. The developers say they "had to modify" that suggestion "extensively" before it became the 10 lines of code that shipped. Ars notes that at least one social media user has said the bug could have been fixed with significant human effort.
The studio did not disclose the AI use when Big Walk launched on Steam earlier this year. It says Valve "does not request disclosure of code or chatbot usage." Valve's disclosure form for developers is aimed at AI-generated content and excludes "efficiency gains" from development environments with built-in AI tools. The IGF makes no such distinction, so House House mentioned the bug fix when it submitted the game. It says it is now making the disclosure public because "it has become a matter of public interest."
House House framed the statement carefully. "This is not an endorsement of, or argument for, the use of AI tools, however minimal," the studio wrote. It said it worried that voluntary disclosures could contribute to the technology being treated as "inevitable," and apologised that its use "may not meet" the standards of those trying to limit harm from generative AI.
Judges divided
Some of the strongest reactions came from jurors who admired the game. Austin Walker called Big Walk "an achievement" and "beautiful" in a blog post. Damon Reece wrote that it is "seriously special to me." Both said even minimal AI use outweighed that regard for awards purposes. "This is pain that has to happen to try and steer our culture away from this," Reece wrote, adding that to do otherwise "is rank cowardice."
Other judges disagreed. Juror Liz Ryerson wrote that the studio's AI use was not personally disqualifying, and was bothered that many "acted like there aren't any levels to this." Ryerson also objected that developers who are transparent about small uses can be punished more than those who disclose nothing.
Outside the jury, views were similarly split. Dominic Tarason, a Steam Hype staff writer and curator, called Big Walk "very mildly compromised art" that is still "very human." AI researcher Mike Cook wrote that while House House is not replacing developers with AI, its actions are "part of the same problem." Game developer and professor Brendan Keogh argued that zero tolerance is "ultimately counterproductive" because anger at conflicted developers who disclose encourages heavier users to hide their use. Narrative designer Hannah Nicklin defended zero tolerance as "a viable political position" aimed at a system rather than at individuals.
The reaction has reached players too. Ars reports that Big Walk's Steam page received more than 400 negative reviews in a few days, though Steam's review-bomb protections appear to be limiting the effect on its "Very Positive" rating.
There is a recent precedent. Last year the Indie Game Awards, a separate event from the IGF, rescinded awards for Clair Obscur: Expedition 33 after its developers admitted some in-game textures had been made with AI tools. Those textures were later patched out.
What it means for work
The task at issue here is narrow: diagnosing an engine bug that a vendor engineer had called unfixable. That is exactly the kind of work coding assistants are now used for, and Cook's point is that it once might have gone to a hired expert. For small studios, the case shows that disclosure rules now differ by venue. Valve exempts coding assistance, while some festival jurors count any use. As Ars notes, AI-assisted code is increasingly likely to appear inside the engines studios build on, which may make a zero-use standard hard to verify or meet. What remains open is whether festivals will define tiers of use or adopt the blanket ban the jurors requested, and whether either approach makes developers more or less willing to disclose.
Sources
This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.
More from māchine
All storiesThe FAA's $875 million SMART system now gives routing and sequencing advice across all US airspace. Airlines back it. The controllers' union says it was never consulted.
Google says its new model scores above its larger Gemini 4 Argon on one index and costs $1.50 per million input tokens. It is meant to carry out the work while bigger models do the planning.
Nvidia's board accepted Huang's resignation on Wednesday. CFO Colette Kress is now interim CEO, and the company has no clear successor as US-China tensions over AI chips grow.