Google Gemini Exposes Gmail and Drive to Human Reviewers

Google reversed its 2023 commitment that personal content from Gmail, Docs, and Drive would never be seen by human reviewers, rewriting its Gemini Apps Privacy Hub on October 16 without a press release or user email, according to a GitHub project documenting the change that reached the top of Hacker News with 366 upvotes. The updated policy states that "a subset of Connected Apps data processed by Gemini is reviewed by human reviewers, including service providers," with reviewed conversations retained for up to three years even after users delete their Gemini activity history. The project cites AI Weekly reporting that file ingestion happens autonomously in background conversations without explicit @Drive or @Gmail tags, and cites Verizon's 2026 DBIR finding that 67% of users access AI on corporate devices through personal, non-corporate accounts that bypass Workspace enterprise protections.

Last week, a GitHub project documenting Gemini File Leaking Behavior https://github.com/thisisme6598/gemini files/ hit the top of Hacker News with 366 upvotes after revealing that Google had silently reversed its 2023 commitment that personal content from Gmail, Docs, and Drive would never be seen by human reviewers. The policy update happened on October 16 — no press release, no email to users, just a rewritten paragraph in a help article. If you use Gemini with Connected Apps enabled and Keep Activity on the default , your emails and Drive files are now being routed to human contractor review pipelines and stored for up to three years — even after you delete them. Google’s Broken Promise In 2023, when Google launched Gemini’s Connected Apps feature, the company was explicit: personal content from Gmail, Docs, and Drive “is not seen by human reviewers.” That pledge was a selling point. It’s why enterprise teams felt comfortable letting employees connect their work accounts to Gemini. Three years later, Google quietly rewrote that commitment. The updated Gemini Apps Privacy Hub now reads: “a subset of Connected Apps data processed by Gemini is reviewed by human reviewers, including service providers.” Reviewers receive summaries and excerpts from your emails and files, generated inferences about your content, and in some cases, the complete unredacted text. Reviewed conversations are retained for up to three years — data that survives even after you delete your Gemini activity history, with no clear erasure pathway. Google’s own privacy page now warns users: “don’t enter anything you wouldn’t want a human reviewer to see.” That is a remarkable sentence from the company that spent three years assuring you the opposite was true. The Part That Makes This Worse The file access isn’t limited to moments when you explicitly ask Gemini to look at your Drive. According to reporting from AI Weekly https://aiweekly.co/alerts/google-gemini-routes-personal-drive-gmail-to-human-reviewers , ingestion happens “autonomously in background conversation without explicit @Drive or @Gmail tags.” You don’t have to @-mention a file for it to be pulled into context. The capture is also probabilistic — identical queries can produce different results across sessions — which means there is no reliable way to audit what triggered a file retrieval or which conversations were flagged for human review. You are flying blind on what a contractor has seen. The Enterprise Risk Nobody Is Talking About If you work at a company using Google Workspace, you might assume you are protected. Workspace enterprise accounts do operate under a Cloud Data Processing Addendum that explicitly prohibits human review for model training outside your domain. That protection is real — but only for content accessed through your company Workspace account. Here is the problem: Verizon’s 2026 DBIR https://jazz.security/dlp-explained/dlp-basics/is-gemini-safe-for-company-data/ found that 67% of users access AI on corporate devices through personal non-corporate accounts. A personal Gmail tab in Firefox on your work laptop bypasses every enterprise control entirely. Google’s admin tools only govern the managed company account. The personal Gmail session next to it is fully exposed to consumer review terms, and IT has no visibility into it. For developers building on Gemini, the exposure compounds. The free Gemini API and Google AI Studio fall under terms that allow Google to “read, annotate, and process your API input and output for product improvement.” If you are using a personal account for Code Assist or AI Studio while building your app, your code is in scope. The fix is straightforward — move to a billed Cloud project, which applies the Data Processing Addendum — but many developers do not know this distinction exists. How to Opt Out To opt out, disconnect Connected Apps in Gemini: open the app, tap your profile picture, go to Personal Intelligence, and disconnect Gmail, Drive, and other apps. On desktop, it is under Settings. TechPP has a step-by-step guide https://techpp.com/2026/10/26/google-personal-intelligence-opt-out/ for each platform. The tradeoff is real — you lose Gemini’s personalized features and contextual memory. Alternatively, disabling Keep Activity entirely is more restrictive but also eliminates Gemini’s ability to retain context across sessions. Neither option is clean. Google designed a system where full privacy requires you to give up the core functionality. Enterprises should enforce Workspace accounts through managed Chrome browsers and implement endpoint detection to catch personal Gemini usage in any browser. This is a wider pattern in AI security — the need to control what agents and tools access on behalf of users is becoming a core enterprise concern, something tools like CrowdStrike’s AI Kill Switch https://byteiota.com/crowdstrike-ai-kill-switch-stops-rogue-agents-dead/ are specifically being built to address. The Takeaway Ultimately, Google breaking a privacy pledge it made in 2023 with a silent blog edit is not a policy clarification — it is a breach of trust with no accountability. The “you agreed to the Terms of Service” defense does not cover explicit, specific promises that users reasonably relied on when deciding to connect their email and files to an AI system. Moreover, if you are building on Gemini, understand what your users are consenting to — and make sure your app surfaces it clearly. And if you are a developer or IT professional advising on AI tool adoption, this is precisely why privacy commitments need to be in contracts, not blog posts.