Is Google Gemini trained on Google docs? One indie developer thinks so, after it told players about his unreleased game plans
AI-summarised brief · reviewed before publication
An independent game developer alleges that Google’s Gemini AI accessed private, unreleased details from their Google Docs without explicit permission. The developer claims a player used Gemini to learn specific information, including the character name “Vantage Tripod” and unreleased mechanics, which had only been written in a private document the day prior. While the public knew of a “Tripod Fish” character, the exact name remained secret until the AI interaction. Google maintains that Gemini only accesses Google Docs with express user permission, such as for summarization, and handles data transiently without retention. The company notes exceptions where publicly linked documents or third-party extensions might expose data. Despite these assurances, the developer and some users remain skeptical, citing the accuracy of the leaked information as evidence of unauthorized scraping. Privacy advocates point to ambiguities in Google’s privacy hub regarding content usage for AI training, fueling concerns about data security and potential unauthorized access to private user files within the Google ecosystem.
💡 Why It Matters
- · This incident exposes a critical trust deficit in AI data handling, where perceived privacy guarantees clash with user experiences of data leakage.
- · It forces a reevaluation of how cloud-based AI models interact with private user content, challenging the assumption that "private" documents are safe from training data ingestion.
