Google Gemini Allegedly Leaked Unreleased Game Content From Private Google Doc

Google Gemini Reveals Unreleased Game Content

Klub Kofta Studio, the solo indie developer behind the tower defense Steam game Operation Octo, has raised alarm after Google Gemini disclosed confidential game content that allegedly existed only in a private Google Doc1

1

. The incident began when a player in the game's Discord server was asking Google's AI search systems silly questions about Operation Octo to get amusing responses2

2

. However, the AI-generated answers contained surprisingly obscure information that should not be available for a small indie game with limited online presence.

How Google's AI Somehow Knew a Game Secret

When the developer asked the player to query Google Gemini about unreleased game content, the AI disclosed the exact character name "Vantage Tripod"—a highly specific detail the developer insists was never mentioned to anyone3

3

. According to Klub Kofta Studio, this character name existed digitally only in one of their private Google Docs and was not present in the game's code, files, or Steam data2

2

. The developer acknowledged that during a game of Gartic Phone, fans had called a related element the "Tripod Fish," but the specific name "Vantage Tripod" had never been shared publicly1

1

. Beyond this character reveal, AI search leaked unreleased game content about mechanics the developer had written into Google Workspace documents just the day before the player's interaction4

4

.

AI Data Training Privacy Concerns Emerge

The incident has sparked questions about whether AI trained on Google Docs could be accessing user data without explicit consent. Klub Kofta Studio believes Google scraped their confidential game content from private documents to train its AI models1

1

. This raises significant AI data training privacy concerns for developers and users storing sensitive information in Google Drive files. Google denies these allegations, stating it does not use private Workspace content, including Drive and Docs, to train its foundation AI models, including Google Gemini2

2

. The company explained that publicly shared documents with links posted online can be indexed by search engine crawlers, but insisted private documents remain protected1

1

.

Privacy Loopholes and User Data Access

However, the situation is more complex than it appears. Google automatically opts users into sharing their Google Drive data with Google to improve AI models, among other uses5

5

. Users on Reddit and other platforms have pointed out that Google Gemini can be automatically connected to Workspace, including Docs and Drive, unless users manually toggle off this integration through Settings > Connected apps1

1

. This default opt-in approach has fueled privacy concerns about AI accessing private data without users fully understanding the implications. Privacy company Proton has noted that Google's privacy hub does not explicitly state it won't use user content for AI training4

4

.

The Mystery Deepens

The available evidence does not definitively establish how Google's AI produced the character name. The original response could not be independently recreated after the Reddit post went viral2

2

. Once the claim appeared in widely shared posts, subsequent queries to Google Gemini began returning the name "Vantage Tripod" while citing the Reddit thread as the source2

2

. This makes it difficult to separate the original response from what the AI can now retrieve from the public web. The developer acknowledged that their specific case is no longer replicable in its original form2

2

.

Unintended Consequences of AI

This incident highlights the unintended consequences of AI and raises questions about how AI search systems handle obscure information. Google AI Overview has a documented habit of pulling information from Reddit threads for answers, correct or otherwise5

5

. The system can also be manipulated—last year, a YouTuber convinced Google's AI search that Grand Theft Auto 6 would include a "twerk button" through a low-effort project2

2

. For developers like Klub Kofta Studio, whose Operation Octo launched on Steam in September 2025, the case serves as a reminder that unreleased project information can be difficult to contain once it reaches a public AI answer or online discussion2

2

.

What This Means for Developers and Users

The data leak raises fundamental questions about data scraping practices and whether users can trust that their private documents remain confidential. Some commenters have speculated about more serious implications: "Doesn't that mean you can dox people by asking very specific questions given that people's private personal data is included in the training sets then?" one Reddit user wondered5

5

. For indie developers working with limited resources, the possibility that AI models could inadvertently reveal confidential game content adds another layer of complexity to game development. Watch for how Google responds to mounting pressure to clarify its data handling practices and whether the company will implement stricter safeguards to prevent AI accessing private data without explicit user permission.