Valve Testing 'SteamGPT'

Valve appears to be testing an internal artificial intelligence system called SteamGPT, based on code references found in a recent Steam client update. (pcworld.com) The references surfaced after an April 7, 2026 Steam update, when Valve watcher Gabe Follower posted screenshots of strings that mention “SteamGPT” and related services. SteamTracking, a GitHub project that logs changes to Steam files, is one of the places watchers use to spot those additions. (pcworld.com) (github.com) The code points to back-end jobs such as task creation, labeling, summaries, testing, and fine-tuning, which looks more like an employee tool than a public chatbot. Separate references described a “SteamGPTSummary” service tied to account details such as Steam Guard status, fraud flags, country, phone data, playtime, and Valve Anti-Cheat status. (videocardz.com) (arstechnica.com) Reports based on those strings say Valve may be aiming SteamGPT at two jobs first: handling parts of Steam Support and helping review suspicious behavior in Counter-Strike 2. Ars Technica reported that the files suggest a security-review role, while other outlets tied the same strings to support workflows and anti-cheat review. (arstechnica.com) (gadgets360.com) That would fit a platform the size of Steam, which Valve says serves developers through Steamworks and handles store, community, and support systems across the service. An internal triage tool could sort tickets or flag risky accounts before a human reviewer makes a decision. (partner.steamgames.com) (arstechnica.com) Valve has already spent the past two years building policy around artificial intelligence on Steam’s public storefront. In a January 10, 2024 Steamworks post, the company said developers can ship games with pre-generated or live-generated artificial intelligence content if they disclose it and explain safety guardrails for live systems. (store.steampowered.com) (partner.steamgames.com) The current Steamworks content survey still asks developers to describe how artificial intelligence is used and, for live-generated systems, what controls prevent illegal content. That means Valve is not approaching artificial intelligence as a new issue in 2026; it already has review rules in place for games sold on Steam. (partner.steamgames.com) What is new here is the target. The leaked strings point inward, toward moderation, support, and trust-and-safety work inside Valve, not outward toward a feature Steam users can click on today. (videocardz.com) (pcworld.com) There are still basic gaps. Valve has not announced SteamGPT, has not described what model it uses, and has not said whether any system is live beyond internal testing, so the current picture comes from file names and code strings rather than a product launch. (msn.com) (pcworld.com) Until Valve says more, SteamGPT is best understood as a leaked internal project name attached to support and security tooling inside Steam. The next real update will be an official comment, a visible client feature, or another Steam build that shows where the experiment is headed. (pcworld.com) (arstechnica.com)