Highlights
- A solo developer from Klub Kofta Studio alleged that Google's AI leaked a secret unannounced character named Vantage Tripod from his game.
- The developer claims the unreleased character name existed only in a private Google Docs file, raising major privacy and data security concerns.
- Google officially denied scanning private Workspace content to train foundational AI models, stating that only publicly shared and indexed links can be referenced.
Video game developers of any size already face a mountain of challenges, from mass refunds on short games to the nightmare of two studios releasing games with the exact same name during the same week. But a new, deeply modern obstacle might be joining that list: AI scraping private documents and leaking unannounced information. That is exactly what the solo developer behind Klub Kofta Studio alleges happened to him after Google's AI inexplicably revealed a secret character from his tower defense game to a fan.
The baffling situation unfolded in the Discord server for Operation Octo, which launched on Steam in September 2025. According to a post by the developer on Reddit, a player was using incognito mode to ask Google's AI "silly questions" about the game just to get a kick out of the answers. However, the developer noticed that the chatbot was surfacing surprisingly obscure information that shouldn't have been readily available for a small indie title.
Curious, the developer suggested the player ask the AI about future, unannounced content. In a scarily accurate response, the AI spit out the exact, highly specific name of an upcoming character: Vantage Tripod. The developer and the fan shared their befuddled reactions in the Discord chat, completely stunned by the eerie leak. What makes this situation so alarming is just how closely guarded that name was. Klub Kofta Studio confirmed that the phrase "Vantage Tripod" had never been shared publicly, nor was it hidden anywhere in the game's code, local files, Steam data, or serialized data.
The developer insists that, to their knowledge, the exact phrase "Vantage Tripod" existed digitally in only one place in the entire world: a private Google Docs file. Because the game has a relatively small online footprint, the creator suspects the AI wasn't confused by online speculation and somehow pulled the real leak directly from their private workspace.
Google Responds and Rebuts the Claims
Google has officially pushed back against the developer's implications. In a statement to Polygon, a company representative clarified that Google does not scan private Workspace content, which includes Google Drive and Docs, to train foundational AI models like Gemini.
The tech giant offered additional context, noting that links to publicly shared documents might be indexed by search engine crawlers, but only if someone actively posts those links somewhere in the public domain. Google also emphasized that users are in full control over their sharing settings for Drive content, pointing to their privacy resources for those looking to adjust their permissions.
If you try to replicate this exact interaction today, you are out of luck. Now that the developer's Reddit post has gone viral, asking the AI about the unreleased character simply results in it citing the Reddit thread itself as its source.
This creates a frustrating loop, but it isn't out of character for Google's AI overviews, which have a well-documented habit of pulling information, correct or otherwise, directly from Reddit threads. The AI overview can easily be manipulated into doling out unreliable misinformation. Just last year, a YouTuber managed to convince Google's AI search that Grand Theft Auto 6 would include a "twerk button" through a fairly low-effort shitposting campaign.
Whether this was an impossibly lucky guess based on the "Tripod Fish" pattern-matching or something deeper, the incident has sparked widespread privacy concerns across Reddit and Bluesky. For years, massive tech companies have been quietly updating their terms of service, often automatically opting users into sharing their data to improve AI models.
Gamers and developers alike are expressing their frustration. One Reddit user lamented how companies quietly opt users into AI training while burying the option to opt out, while another mused about the terrifying potential of people being doxxed if their private personal data is unknowingly swept up into AI training sets.
