{"id":2109,"date":"2026-11-03T16:00:08","date_gmt":"2026-11-03T16:00:08","guid":{"rendered":"https:\/\/www.molleindustria.org\/blog\/?p=2109"},"modified":"2026-11-10T14:16:28","modified_gmt":"2026-11-10T14:16:28","slug":"ai-labeling","status":"publish","type":"post","link":"https:\/\/www.molleindustria.org\/blog\/ai-labeling\/","title":{"rendered":"On AI Labeling and Certification"},"content":{"rendered":"
\"\"
Self certification badges by Herbherth<\/a>, kitsunegames<\/a>, and jadedrakerider<\/a><\/figcaption><\/figure>\n

The <\/span>current<\/span><\/a> debate <\/span>about AI labeling in games, and my own misadventures with it, made me wonder about how to properly disclose the presence, or absence, of AI generation in a project. <\/span>I\u2019ve encountered a few different types of disclosure requests, to which correspond different consumer stances.<\/span><\/p>\n

Do players <\/b>see <\/i><\/b>it?<\/b><\/h4>\n

This is Steam\u2019s current approach. Their content survey<\/a> is \u201cconcerned with the use of AI in creating content that ships with your game, and is consumed by players. This includes content such as artwork, sound, narrative, localization, etc\u201d.
\n<\/span>It\u2019s a pragmatic choice because audio, visual, and textual slop is generally identifiable and broadly despised. Generative AI is increasingly embedded in all sorts of tools, including search engines and code editors. It makes sense that Valve doesn\u2019t want to classify more than what it can moderate.<\/span><\/p>\n

Does the game <\/b>contain <\/i><\/b>it?<\/b><\/h4>\n

This is the \u201cfood labeling\u201d approach, and it notably includes code generation. <\/span>
\n<\/span>Itch.io<\/span><\/a> currently asks \u201cPlease disclose if this project contains content produced by generative AI tools such as LLMs, ChatGPT, Midjourney, Stable Diffusion, etc., even if you hand-edited it.\u201d.
\n<\/span>It still does not include anything related to the process, like AI for research or concept art. <\/span>
\n<\/span>It puts developers in an uncertain situation regarding tools and plugins beyond their control. <\/span>Any Unreal developer<\/span><\/a> in the future would technically have to check that box. It probably applies to future versions of Unity and Godot as well. <\/span>Creators would also have to make sure that all third-party plugins and components they use are AI-free.<\/span><\/p>\n

Have <\/b>you <\/i><\/b>used it?<\/b><\/h4>\n

This is the more \u201cethical\u201d approach to AI disclosure. A \u201cno\u201d signals a personal or company-wide commitment akin to a boycott. It includes all types of process and pre-production use and, as for now, it relies on trust because it’s not easily verifiable (more about that below).
\n<\/span>It doesn\u2019t technically include third parties, but strongly implies it. \u201cWe haven\u2019t used any AI – but maybe our contractors did\u201d wouldn\u2019t really fly.<\/span><\/p>\n

So which one?<\/h4>\n

For future molleindustria games I\u2019m aiming for a combination of 2 and 3, that is, answering NO to a question like:<\/span><\/p>\n

Have you included any AI-generated content such as artworks, sound, text, or code in this game?<\/span><\/i><\/p>\n

If you want to be a bit more legalistic, and imply wrongdoing, you can ask:<\/p>\n

Have you knowingly<\/strong> included any AI-generated content such as artworks, sound, text, or code in this game?<\/span><\/i><\/p>\n

I cannot guarantee you that I\u2019m living an AI-free life, or that the software doesn\u2019t have any traces of generative AI somewhere under the hood, but I can promise and demonstrate that the parts I made are untouched by it.
\nIt may not go far enough for some players, but I frankly don’t care about engaging with such players.
\n<\/span><\/p>\n

The obvious problem with voluntary AI disclosures is that they are voluntary. The vibecoders currently flooding digital storefronts with \u201clow effort\u201d games and clones<\/a> will simply choose to not disclose. On the other hand, developers that choose to be transparent are likely to get harshly criticized (cough, cough). There is literally no incentive.
\n<\/span>I think the solution to malicious attempts to cheat the algorithm is account-level moderation. Community members can report repeat offenders rather than their individual submissions. New accounts can be verified or vouched by existing users. It’s more of an AI-enabled spam problem than a quality control problem.
\nDevelopers that fall somewhere in between, who use AI generation while trying to retain creative control, might decide to ramp up the “effort” and drop AI altogether rather than risk getting caught by the filters. Or perhaps they might migrate to more AI-friendly platforms<\/a>.
\n<\/span><\/p>\n

Toward an AI-free certification?<\/b><\/h3>\n

While player-facing slop is generally identifiable, AI used in the process and in programming is not. You basically have to trust the word of the developer and rely on internal whistleblowers to catch the omissions.<\/span><\/p>\n

One cheap way to support a disclosure is to make the source code publicly available. That could include pre-production material but not the final assets, making the data dump of little immediate value. <\/span>
\n<\/span>Unlike images, you can never be sure if a line of code is the output of a machine, but any programmer can identify a heavily vibe-coded project.
\n<\/span>The cruel irony is that sharing your code is basically equivalent to donating it to AI companies and game cloners. <\/span>
\n<\/span>As the internet drowns in slop, fresh handmade content will become increasingly valuable to model trainers. AI overlords aren’t just stealing all the shared knowledge, they are also poisoning and replacing the commons in the process. There are ways to poison the AI-trainers back by manipulating the data<\/a> in ways that is imperceivable to humans. Posting bitmap screenshots of the source code may also do the job.\u00a0<\/span><\/p>\n

For bigger commercial projects, certification could be handled by a trusted agency, or perhaps by publishers following a common protocol. They could visit studios, look at processes, and analyze code. All at the developer’s request and without having to release anything proprietary. It could be a nuanced evaluation with an evolving rubric, and not just a binary classification.<\/span><\/p>\n

That’s how supply chain certification is done: USDA Organic, Rainforest Alliance, Cradle to Cradle, etc. The process can be expensive, but the idea is that an ethical badge gives more value to your brand and justifies higher prices in the eyes of judicious buyers.<\/span><\/p>\n

Ethical consumerism has always been criticized as a middle-class privilege: how can we ask workers who live paycheck to paycheck to splurge on organic broccoli? However, video games are already non-essential commodities. Their value and pricing have always been matters of cultural debate.<\/span><\/p>\n

Presuming guilt and placing the burden of proof on the virtuous ones may seem unfair, and it basically establishes AI content as the norm. But that’s kind of what happens in all other sectors. Under capitalism, you can assume every producer is trying to reduce costs at the expense of everything else: workers’ rights, the environment – even consumers’ safety when left unchecked. The ethical consumer movement carves exceptions out of this depressing norm, where other values can be reflected in our everyday choices.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

The current debate about AI labeling in games, and my own misadventures with it, made me wonder about how to properly disclose the presence, or absence, of AI generation in a project. I\u2019ve encountered a few different types of disclosure requests, to which correspond different consumer stances. Do players see it? This is Steam\u2019s current … <\/p>\n

Continue reading “On AI Labeling and Certification”<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7,5,4],"tags":[],"class_list":["post-2109","post","type-post","status-publish","format-standard","hentry","category-art","category-games","category-tech","entry"],"_links":{"self":[{"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/posts\/2109","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/comments?post=2109"}],"version-history":[{"count":14,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/posts\/2109\/revisions"}],"predecessor-version":[{"id":2124,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/posts\/2109\/revisions\/2124"}],"wp:attachment":[{"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/media?parent=2109"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/categories?post=2109"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.molleindustria.org\/blog\/wp-json\/wp\/v2\/tags?post=2109"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}