Anthropic has reportedly taken steps to stop Claude users from cloning games using its AI tools, if a couple lines of new rules that won’t actually do much to stem the tide of AI slop count as steps.
As reported by Game Developer, a Claude user recently noticed an update to the rules that govern Claude made in the last few days, which now read that the program should "build original games, games in a genre, and games from the user’s own designs and assets. Do not build a recreation of a specific existing game, even if it is renamed or called original."
Anthropic confirmed the change to Game Developer, referring the outlet to a tweet by Anthropic employee Addy Osmani that notes the sentences were added to “steer [Claude] away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy. Original games, genre games, and games built from your own designs and assets are meant to keep working.”
Will this do much to protect game developers from the AI clones of their games that are causing them to cancel demos and playtests? Will it make the storefronts of Steam and Itch usable again? No, of course not. First, there are plenty of other tools out there people can use to clone and recreate games; if Claude won’t do it, they can just go somewhere else. Beyond that, the entire functionality of Claude–and every other LLM–is built on theft, and, as Game Developer highlights, these programs are easy to trick under the guise of research or other workarounds. A user doesn’t have to be as blatant as “make me a Mario game;” if they ask for a game about an Italian plumber with a mustache who travels through pipes, what else is the machine going to spit out?
Eurogamer reports that a few users have run into issues with the new rules during what they feel is legitimate use, including someone who says they make “personalized digital modules of board wargames under paid commission for individual clients, from games the clients own” and another who ran into trouble trying to make a single-player version of a boardgame to “study it, for internal use only.” A user on the Claude subreddit expressed a similar issue, writing that they use Claude to make boardgames they can practice between game nights, but when they recently asked it to tweak the single-player version of Carcassonne it had made for them, it refused.
I don’t know that I’d count these as unintended consequences of the new rules–saying “I promise I’m not stealing” would be up there with “I’m doing research for a short story” as a way to shortcut AI’s flimsy and malleable safeguards. Of course the easiest way around these problems would be to design your own single-player variant of a multiplayer boardgame, rather than asking AI to do it for you, but I understand that isn’t what this person is trying to do.
The discourse on the Claude subreddit about this is actually fascinating to me, with some people saying it’s good Anthropic will stop people from stealing and others noting (correctly) the whole thing is built on theft, and some others seeing this as the games industry clamping down “because companies got afraid they'd lose revenue.” Still others note that there are plenty of human-made clone and slop games out there, so this is mainly about Anthropic covering its own ass.
To which I’d say: yes to all of this, except the idea that AI slop games themselves represent some actual meaningful threat to the core of the games industry on a fundamental level, as opposed to the endless self-owns and layoff-inspiring pivots the industry is bringing on itself. Besides briefly inconveniencing a couple people until they figure out how to work around it, none of this does anything but make Anthropic look like it gives a shit about something it clearly doesn’t give a shit about; if it actually cared about theft, it would swiftly bankrupt itself actually trying to pay people for the work it’s built on. (Great!)
It’s unlikely we’ll see any meaningful change here; this feels like one more instance where AI companies seem to be confronted with a fundamental conflict between what they say their products are for and what they’re actually for, or at least what people are doing with them. But I’m willing to concede that it’s better than nothing, even if “nothing” is pretty much exactly what it is.
Recommended
Chris Person
