Leaked 12TB Steam archive resurfaces a decade of lost PC game builds

A 12TB dump of Valve's old Steam2 servers has surfaced online, exposing pre-2013 prototype builds of Portal 2, GTA 3, Mafia 2 and hundreds more.

AI-drafted from 3 sources. This story was written with AI from the outside reporting listed under Sources, and checked by automated tools before it went up. The originals have the full detail.

Over the weekend a 12 terabyte archive of old Steam server data began circulating on BitTorrent, and it contains more than a decade of PC gaming history that many assumed was gone for good. Dubbed the “teraleak” by the people digging through it, the dump covers Steam's defunct Steam2 content system, which Valve used to distribute games between 2003 and 2013, and it includes not just finished releases but prototypes, playtest builds and cut content from both Valve's own games and hundreds of third-party titles.

Valve switched from Steam2 to its current SteamPipe system in 2013, moving from a proprietary distribution format to standard HTTP file trees so updates could ship as smaller differential patches rather than whole new builds. Gizmodo reports that the switch stranded a huge amount of older content, which had only survived as local installs on ageing machines, if it survived at all. Archivists had long treated many pre-2013 builds as lost media. This weekend's leak suggests that description was, at best, premature.

Why the leak happened, according to the people who found it

Nobody is claiming Valve was hacked. Longtime Valve dataminer Gabe Follower wrote on social media that he had “verified that everything in Steam2 Teraleak was obtained via a publicly accessible endpoint. It's Valve's fault…”, a line quoted in Ars Technica's report. The Spanish-language streamer Scolcer made a similar claim, saying the material came from a site that was “100% accessible to the public... Hidden in plain sight, but with no protection whatsoever.” Neither claim has been independently verified, and it is still unclear whether the archive was pulled from that endpoint recently or assembled over years from private hoards. A readme file bundled with the release offers “warm n good wishes to all hoarders who had stuff from this collection”, which points at the latter. Moderators from the Valve Cut Content Discord told Ars they suspect the depots (Steam's term for a bundle of files tied to a specific game version) sat in private collections for a long time before being merged into one release.

What is actually in the depots

The scale is the story here. Gizmodo dumped the raw depot list into a spreadsheet and got 10,876 lines, though that overstates things slightly since some games span several depots and some depots span many games (one, depot 910, apparently bundles trailers for 2,700 different titles). The preservation group Hidden Palace has already catalogued playable builds including a complete 2009 Portal 2 beta, plus prototypes of Left 4 Dead, Counter-Strike: Global Offensive and Valve's cancelled camera game F-Stop. According to a report on the archive's scope, third-party finds include prototypes of Grand Theft Auto III, Mafia 2, Civilization V, Spec Ops: The Line, Crusader Kings II and Sonic the Hedgehog 4.

The Portal 2 beta is the headline find for Valve fans. It differs substantially from the 2011 release: GLaDOS herself guides players through early test chambers using pre-recorded lines rather than handing that job to Wheatley, and other personality cores play a bigger role. Dataminers have also surfaced a weapon model, sometimes called the Weaponizer, that matches footage from the cancelled Half-Life 2: Episode 3 shown in Valve's own 20th anniversary documentary. There is no playable Episode 3 build, but data trawlers have found files in a directory labelled “ep3”, which Ars notes had also turned up in a 2011 leak.

What is still unresolved

Nobody involved, including Valve, has confirmed exactly how the archive came together, and Valve has not commented publicly. Commentator Tyler McVicker argued the third-party material makes this “a dramatically more dangerous situation than if it were just some Valve builds”, since publishers like Rockstar and Sega now have their old prototypes circulating outside their control. For preservationists the calculus is different: material once written off as lost is suddenly documentable, provided anyone can work out which builds actually run, since some listed depots are already missing required assets. With 12TB still being sifted, more finds, and more awkward questions, seem likely.

Sources

Image: Ars Technica

More news from Gaming World