When a player asked Google’s Gemini AI about an upcoming indie title, the chatbot answered with details the developer says were never public. The response included the exact name of a character—Vantage Tripod—and descriptions of mechanics that the creator had only entered into a private Google Doc the previous day. The developer posted the exchange on Reddit, attaching screenshots that he says prove Gemini accessed his confidential files.
According to the Reddit thread, the game’s only publicly known element was a placeholder called “Tripod Fish.” The AI’s mention of the precise name and specific abilities surprised the creator, who believes the only source of that information was his internal document. He acknowledges that Gemini made a few errors, but the accuracy of the disclosed data led him to suspect the AI had scraped his private Docs.
Google’s official stance is that Gemini can read Google Docs only when a user grants explicit permission—such as asking the model to summarize a document. The company adds that any data the AI accesses is handled transiently and is not retained for future training. Exceptions exist, however, if a document is shared publicly via a “anyone with the link” setting or if a third‑party extension is allowed to read the file. In those cases, the content could be indexed for training purposes.
The developer’s claim raises questions about how Google sources data for Gemini. Some users argue that the tech giant may be harvesting publicly accessible Docs or exploiting extensions that have broader access. Privacy‑focused firm Proton has highlighted that Google’s privacy hub does not explicitly promise that user content will be excluded from AI training, fueling further skepticism.
TechRadar attempted to replicate the incident by prompting Gemini with similar queries, but the model did not reproduce the leaked details. The developer’s friend, who originally triggered the AI response, also failed to recreate the exact answer in subsequent tries. This inconsistency leaves open the possibility that the AI hallucinated the information, drew from prior prompts, or that the developer unintentionally exposed the data through another channel.
Regardless of the cause, the episode underscores the risks of storing sensitive information in cloud services that AI models can access. Users are reminded to review sharing settings, limit third‑party extensions, and treat any data entered into an AI as potentially public. The incident adds to a growing list of concerns about AI privacy and the transparency of training data practices.
Google has not yet responded to requests for comment. Until the company provides a clear explanation, developers and other users will likely remain wary of how their private documents might be used by generative AI systems.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.