Skip to main content

ARS TECHNICA·

Google Gemini Privacy Risks in Gmail and Drive: Explained

9 min listenArs Technica

Google’s Gemini integration into Gmail and Drive raises privacy concerns. Learn why opting out is difficult and how user data influences AI development.

Transcript
AI-generatedLightly edited for clarity.

From DailyListen, I'm Alex

HOST

From DailyListen, I'm Alex. You probably saw the headline this morning about Google's Gemini AI sliding into Gmail, Drive, and Docs as a sidebar helper. It's pitched as a research booster for paying users, but Ars Technica calls out hidden costs in those AI defaults and an illusion of choice on data sharing. Opting out sounds like a maze. Stakes are high with Gemini hitting 2 billion monthly users on AI Overviews alone. We're joined by Priya, our technology analyst, who tracks how these integrations hit everyday workflows and privacy. Priya, what does this sidebar rollout unlock first?

PRIYA

What this unlocks is Gemini pulling from your personal Gmail and Drive data to draft replies or summarize files—right in the sidebar for Google Workspace, Education, or One AI Premium subscribers. Google rolled it out fully within 15 days for most paying users last year. They promise not to use your Workspace data to train foundational models like Gemini itself without permission. But here's the catch: when you use Gemini for those isolated tasks, it processes your emails or docs on the spot. Outputs from those sessions could loop back into broader training datasets indirectly. Gemini's exploded to 750 million monthly app users by early this year, up from 450 million at the start of 2025. That's double the growth in 15 months, powering 2 billion AI Overview views monthly. For busy pros, it speeds research, but it defaults your data into the mix unless you hunt obscure settings.

HOST

Those numbers are wild—2 billion AI Overview users monthly, that's like every other internet user touching it. But you said outputs could influence training indirectly. How real is that risk for someone's actual inbox?

PRIYA

The interesting piece is how Gemini handles data during those "isolated tasks" in Gmail or Drive. Google says it doesn't save your content post-processing within Workspace apps. Yet any interaction sends snippets to their servers for the AI to generate responses. Those generated outputs—say, a summary of your emails—aren't ring-fenced forever. They can feed into aggregated datasets that refine future models. Google insists foundational training skips user data, per their PCMag statement. Still, with 650 million monthly app users logging in via a late 2025 blog post, scale amplifies leaks. Compare to Microsoft Copilot in Office: they explicitly block customer data from training OpenAI models. Google's looser on outputs. For you drafting a client email, one query risks your thread echoing in model tweaks down the line.

HOST

Outputs feeding back sounds sneaky—especially since paying for Workspace or One AI gets you these features by default. Isn't opting out supposed to protect that?

PRIYA

Opting out demands digging through buried menus, often disabling core features like personalized search or backups. Arizona's AG sued last year, claiming Google makes location tracking opt-outs "impractical if not impossible," even when you toggle it off. Similar for Gemini: no simple "Reject All" button like France's CNIL forced post their €150 million fine in 2022. You might lose Drive syncing or Gmail filters. Tom's Guide tester hooked Gemini to their inbox and raved about research speed, but skipped the data toggle hassle. Over 8 million paid Gemini Enterprise seats across 2,800 companies by early 2026 show businesses biting anyway. It's choice, but rigged toward on.

That CNIL fine hit €150 million just to add a Reject...

HOST

That CNIL fine hit €150 million just to add a Reject All—yet here we are with Gemini defaults. How does this tie into Google's bigger defaults playbook?

PRIYA

Defaults lock in behavior. EU's 2018 choice screen on Android cut Google's Russia search share from 60% to 40%, but overall rivals barely gained. ProMarket notes users stick to Google 90% of the time post-choice. Gemini mirrors that: it's baked into Search as AI Overviews for 2 billion users monthly, no opt-out without URL hacks like udm=14. Antitrust suits hammer this—federal judge ruled last year Google monopolized search via default deals on Apple and Android. Adtech case added monopoly in publisher servers and exchanges; one rival lost 40% revenue, laid off 45% staff. Gemini's 35 million daily actives mean billions of queries default to AI processing your data. Change the default, and choice emerges.

HOST

Russia dropped to 40% after the screen—that's real dent, unlike EU-wide. But Gemini's at 750 million users. Aren't people voting with their feet here?

PRIYA

People pick Gemini because it's free in Search and Workspace upsells convenience. But defaults create inertia. Google's antitrust defense in Reason.org piece argues users overwhelmingly choose them anyway—EU screen proved it minimally shifts shares. Yet plaintiffs counter exclusive defaults on billions of devices bar rivals. Gemini Nano to Ultra models power it all, from phone summaries to enterprise. Early 2026 stats: 4.3 million app downloads by November 2025, now 650 million monthly actives. Growth's real, but tied to Gmail's 1.8 billion users auto-exposed. Fortune reports Gemini now scans emails and photos for "Personal Intelligence." Without nudges, few switch—mirroring how Bard became Gemini, dominating chatbots.

HOST

Personal Intelligence scanning photos too? With those antitrust rulings piling up, does this data grab fuel breakup talks?

PRIYA

Antitrust spotlight intensifies the data angle. Judge ruled Google illegally tied ad server DFP and exchange AdX in Google Ad Manager, killing rivals—one CEO emailed in 2019 that nearly every publisher server folded. Internal Google note called switching "an act of God" due to costs. Search monopoly ruling last year spotlighted defaults; remedies could split ad business. Gemini integrations amplify: 2 billion Overviews users feed ad targeting. NBC News caught hallucinations like "elephants have two feet" or Obama as Muslim president—viral fails erode trust. 77% of businesses worry per surveys, as advanced AI hits 60-85% accuracy versus human 95%+. Defaults hide these costs.

Hallucinations claiming Obama was Muslim president

HOST

Hallucinations claiming Obama was Muslim president? That's not just wrong, it's dangerous. How do defaults make those risks stickier in Gmail or Drive?

PRIYA

Defaults embed hallucinations into trusted tools. Gemini sidebar in Docs might summarize a report with fabricated facts—say, wrong sales figures from your Drive files. Outputs aren't saved per Google, but errors propagate if you share them. NBC reproduced "eat rocks" advice from Overviews; social media mocks #GoogEnough. In translation, businesses pay double: AI then human checks, since models confabulate 15-40% on complex tasks. Consensus from multiple models cuts that, but Google's solo default skips it. With 35 million daily Gemini users, one bad summary in your inbox cascades to clients. People + AI Guidebook from Google's research urges weighing long-term user hits over quick automation.

HOST

Double-paying for fixes on top of defaults—brutal for pros relying on Gmail daily. Google claims they respect privacy. Where's the gap there?

PRIYA

Google's spokesperson told PCMag they skip Workspace data for training outside without permission—an August 2025 post reiterated it. Gap's in practice: class action in California alleges Incognito mode still tracks you, violating Wiretap Act. Arizona suit says location pings persist post-opt-out. Gemini processes data server-side for every sidebar query, even if not stored. Compare OpenAI: ChatGPT Enterprise blocks training entirely. Google's 277-page adtech ruling exposed no edge on fraud versus rivals. For your Drive folder, one "summarize" click risks exposure without clear audit trails. Rollout to 8 million enterprise seats shows tolerance, but lawsuits mount.

HOST

Incognito not incognito—echoes those suits. If defaults are the illusion, what's next for users dodging this?

PRIYA

Users add udm=14 to searches to kill Overviews, or toggle per-app data sharing in Workspace admin panels—tedious for individuals. Enterprise admins at those 2,800 firms lock it domain-wide. Forward: antitrust remedies might force choice screens for AI features, like Russia's search drop. Google pushes generative AI as future, but People + AI notes AI solves specific pains, not all—defaults regress experience if hallucinations hit workflows. Rivals like Anthropic's Claude Enterprise offer data isolation upfront. Gemini's growth to 750 million locks ecosystem, but regs could pry it open. Check settings now; defaults win if you don't.

Choice screens might finally give breathing room

HOST

Choice screens might finally give breathing room. Priya, always eye-opening on these tech shifts. I'm Alex. Thanks for listening to DailyListen.

Sources

  1. 1.Gemini Users Statistics (2026) – Growth & Revenue Data
  2. 2.Google Gemini Statistics 2026: Users, Revenue & Growth
  3. 3.Google Gemini Stats 2026 – Market Share, Users and More. - fatjoe.
  4. 4.Gemini AI Comes to Google Drive, Docs, Gmail as Sidebar | PCMag
  5. 5.I connected Gemini to Gmail and it changed how I research | Tom's Guide
  6. 6.Google's Data Collection Practices Face Scrutiny in Recent Lawsuits | Privacy & Security Law Insights | Davis Wright Tremaine
  7. 7.Privacy concerns with Google - Wikipedia
  8. 8.The hidden cost of Google's AI defaults and the illusion of choice
  9. 9.Google connects Gemini to users' emails and photos in push to build ...
  10. 10.Judge rules Google illegally monopolized adtech, opening door to potential breakup  | TechCrunch
  11. 11.What We Learn About the Behavioral Economics of Defaults From the Google Search Monopolization Case - ProMarket
  12. 12.'Google Is a Monopolist,' Judge Rules in Landmark Antitrust Case
  13. 13.Google AI Overviews under fire for giving dangerous and wrong answers
  14. 14.Worst Google AI Responses Overview
  15. 15.Google news: found guilty of antitrust violations in online ad case.
  16. 16.When AI Hallucinates: The Hidden Cost of Confident Mistakes | AtData
  17. 17.The Cost of AI Hallucinations: A Calculator for Global Business
  18. 18.Google’s AI faces social media mockery after viral errors
  19. 19.One of the flaws in DOJ's anti-trust case: People overwhelmingly choose Google
  20. 20.Trust Is Scarce: The Hidden Cost of AI Hallucinations - Tom Spencer
  21. 21.Identify user needs & AI strengths - People + AI Research
  22. 22.User Needs + Defining Success
  23. 23.Google Forcing Users To Use Its Overview Despite It Not ... - YouTube

Original Article

The hidden cost of Google's AI defaults and the illusion of choice

Ars Technica · April 30, 2026