The developers behind the popular digital gaming storefront Steam have implemented a new set of guidelines that govern the use of AI-powered development tools in new titles released on the storefront.
In a new post on the Steamworks development blog, Bellevue, Wash.-based Valve Software announced it had changed its policies on AI content for Steam. This will enable the company to release, in its own words, “the vast majority of games that use [AI].”
The plan is that it will update the submission process for new Steam games, where creators are required to disclose the role that AI software has played in “the development and execution of your game.”
As part of that process, Valve draws a distinction between two different kinds of AI use: pre-generated and live-generated. The former applies to any creative content, such as art, music, or code, that was created by generative programs, i.e. Midjourney, while the latter refers to any game that uses generative AI to develop assets or scenarios while the game is running.
Microsoft’s partnership with Onworld.AI, for example, includes the possibility of using AI-powered non-player characters in video games. These could be set up to react uniquely and in real time to the player’s actions, rather than choosing between several pre-programmed reactions. Under Steam’s new rules, an Onworld character would be considered a live-generated asset.
The disclosure process on Steam also now requires developers to state that their AI-powered games do not include “illegal or infringing” content, and that the live version of their game will match their marketing material. Further, any game that was made with or contains AI will be labeled as such on its Steam page, in order to effectively communicate that to consumers.
Valve’s new rules are a reversal of its previous policies on AI tools in game development, which became public knowledge in June. At that point, Valve simply prohibited any game that had been developed using AI from being released on Steam at all, citing the risk of copyright infringement.
GeekWire has reached out to Valve for comment.
Valve plans to double-check on its stated AI restrictions during its internal evaluation of any given game, and that it will require developers to tell Valve what “guardrails” have been implemented to guarantee their live-generated AI will not come up with illegal content.
The repeated emphasis on illegal content comes on the heels of a Stanford study from last month, which found that several generative machine learning models had picked up illegal material, such as CSAM images, over the course of scraping the broader internet for training data.
As a related result, Valve has released a new system on Steam where players can report any illegal material that they might encounter in live-generated AI content. This can be done via Steam’s built-in overlay, which is accessed by pressing Shift+Tab while in-game.
“Today’s changes are the result of us improving our understanding of the landscape and risks in this space, as well as talking to game developers using AI, and those building AI tools,” Valve wrote. “This will allow us to be much more open to releasing games using AI technology on Steam.”
The exception is any game that contains “Adult Only Sexual Content that is created with Live-Generated AI,” which is still prohibited on Steam. For example, this would include games that use AI models to create a virtual interactive partner.
