~/AI PRIVACY/google-denies-gemini-trains-on-private-google-docs-after-leak-claim

Google Denies Gemini Trains on Private Google Docs After Leak Claim

Google has denied claims that its Gemini AI trains on private Google Docs after an indie developer reported that the AI leaked unreleased game details stored only in a private document. Google suggested the document's link might have been publicly shared and subsequently indexed by search engine crawlers. This incident highlights growing user anxieties regarding AI data privacy and the potential for sensitive information to be exposed. It also underscores the importance of understanding how sharing permissions and search engine indexing can inadvertently expose private documents to AI models. The leaked information involved an unreleased character named "Vantage Tripod," which the developer claims existed only in a private Google Doc. Google clarified that Gemini only accesses Workspace files when explicitly requested by the user and that users maintain full control over their document permissions.

## BACKGROUND

Google Workspace's privacy policy states that generative AI features like Gemini do not use private customer data to train their models without permission. However, if a Google Doc's sharing settings are set to "anyone with the link" and that link is posted on a public website, search engines can crawl and index the document's content. Once indexed, this information becomes part of the public web data that AI models may access or train on.

## REFERENCES

## KEYWORDS

#AI Privacy#Google Gemini#Data Security#Large Language Models

$ subscribe --daily

Google Denies Gemini Trains on Private Google Docs After Leak Claim | Daily News