Is Google Gemini Safe? We Audited the Privacy Policy, App Permissions, and Legal Record So You Don’t Have To (2026)

Is Gemini Safe?

Google Gemini is safe for everyday casual use — but carries documented privacy trade-offs that casual users miss and that business and enterprise users cannot afford to overlook.

Quick Answer

Google Gemini is a legitimate, registered product of Alphabet Inc. with robust infrastructure security. It is not a scam. However, Axis Intelligence’s safety audit identified four material risks: wide-scope data collection including potential access to call logs and contacts on mobile, human reviewer access to conversations retained for up to three years after deletion, a pending class-action lawsuit (Thele v. Google LLC) alleging unconsented Gmail/Chat/Meet scanning, and a Common Sense Media report labeling it “high risk” for minors. The product is Safe for casual consumer use with privacy settings configured, Caution for users who share sensitive professional information, and Not Recommended without additional controls in regulated-data environments (healthcare, legal, finance).

Overall Safety Verdict: ✅ Safe (with configuration) — ⚠️ Caution for sensitive data users

---

What We Tested

To produce this audit, the Axis Intelligence team reviewed the following inputs between May and June 2026:

1. Privacy Policy review — Full read of the Gemini Apps Privacy Hub (last updated May 19, 2026) and Google’s master Privacy Policy, with key clauses extracted and flagged.
2. App permissions review — Systematic review of Android permissions required by the Gemini app and the host Google app on a Pixel 9 Pro and Samsung Galaxy S25.
3. Signup and onboarding flow — Tested the free and paid subscription flows, documenting what disclosures are made and at what point.
4. SSL and transport security verification — Confirmed HTTPS enforcement and certificate chain validity for gemini.google.com.
5. Payment and subscription terms — Reviewed cancellation and refund policies for Google AI Pro ($19.99/month) and the enterprise billing documentation.
6. User complaint sampling — Reviewed forum threads, Reddit discussions, and documented complaints from Google’s own support community as of June 2026.
7. Legal record review — Reviewed publicly available court filings for Thele v. Google LLC (Case No. 5:25-cv-09704, N.D. Cal.) and the pre-existing Rodriguez v. Google LLC class action.
8. Child safety review — Reviewed Google’s official guidance on minors, the Family Link controls, and a November 2025 Common Sense Media assessment.
9. Support response test — Submitted a billing inquiry through Google One support channels and documented response time.

Axis Intelligence Safety Scoring Matrix

The following scores reflect our assessment based on publicly verifiable evidence as of June 2026. Scores are on a 1–10 scale where 10 = no concerns identified.

Dimension	Score	Key Finding
Privacy Policy Transparency	6/10	Policy is detailed but lengthy; key risks (human review, 3-year retention post-deletion) require active reading to find
Data Collection Scope	4/10	22+ data types collected; mobile app may access call logs, contacts, and screen content via the Google app host
Data Handling & Retention	5/10	18-month default auto-delete; reviewed conversations retained up to 3 years even after user deletes activity
Transport & Infrastructure Security	9/10	TLS/HTTPS enforced on all endpoints; Google’s infrastructure security is among the strongest of any consumer platform
Payment Protection	7/10	Standard Google Pay infrastructure; no prorated refunds on annual plans; partial billing periods non-refundable
Customer Support Quality	6/10	Google One support available for paid users; free tier support is self-service only; no dedicated Gemini safety hotline
Child & Vulnerable User Protections	5/10	Content filters exist for under-18 accounts; under-13 now allowed via Family Link; Common Sense Media rated it “high risk” for kids
Legal & Regulatory Standing	5/10	Legitimate Alphabet company; active class-action (Thele) in discovery phase; separate ongoing Rodriguez appeal
Red Flags / Deceptive Practices	7/10	No outright deceptive patterns found; consent controversy around October 2025 “Smart features” default-on is documented

Composite Safety Score: 6.0 / 10 Methodology: Axis Intelligence analysis. Each dimension weighted equally. Composite is unweighted mean. Scores reflect documented public evidence, not prediction of future behavior.

Risks We Found

1. Broad Data Collection — Including Mobile Sensors

Gemini collects what Google describes as 22 different types of data. This is among the highest counts of any AI assistant reviewed by independent analysts. According to Google’s own Gemini Apps Privacy Hub (updated May 19, 2026), the service collects:

• Everything you type or speak in prompts
• Files, photos, screenshots, and screen content you share
• Transcripts and recordings from Gemini Live sessions
• Location data (general area via IP, or precise location if the device permission is enabled)
• Information about apps on your device (to enable device-assistant functions)
• System permissions and device data when used as the Android device assistant — including call and message logs, and contacts

That last item matters. Gemini runs not as a standalone Android app but as a layer within the Google app, which serves as the host for device-assistant functionality. The Gemini app shortcut inherits all Google app permissions. As Malwarebytes documented in December 2025, this architecture means disabling permissions in the Gemini app listing doesn’t work — they must be revoked at the Google app level under Settings → App Info → Permissions.

What it means in practice: If you use Gemini as your Android assistant and haven’t explicitly audited the Google app’s permission set, Gemini may be accessing more device data than you realize.

2. Human Reviewers Can Read Your Conversations

Google is transparent about this but buries it deep in the privacy documentation. The Privacy Hub states clearly: “Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services.”

More critically: even if you delete your Gemini Apps Activity, any conversation that was reviewed by a human contractor is retained separately for up to three years. Deleting your activity removes it from your visible history — it does not purge the copy held by Google for quality review purposes.

For the unpaid Gemini API tier, Google’s own terms explicitly state that input and output may be used to improve AI products and that human reviewers may “read, annotate, and process” API content, though Google states it disconnects this data from account identifiers before reviewer access.

3. Active Litigation: Thele v. Google LLC

In October 2025, Google allegedly switched on its “Smart features” toggle — which gives Gemini access to Gmail, Chat, and Meet content — for all accounts without a separate, clear consent prompt. A class-action complaint (Thele v. Google LLC, Case No. 5:25-cv-09704, N.D. Cal., filed November 11, 2025) alleges this constitutes a violation of the California Invasion of Privacy Act (CIPA), which prohibits the recording of confidential communications without all-party consent.

As of March 2026, the case was in discovery. Google’s position is that Smart features are disclosed in existing terms and that the toggle was opt-in — but the complaint contends the setting was buried in complex privacy menus not realistically discoverable by average users.

Why this matters for you: If you use Gmail, Google Chat, or Google Meet and haven’t explicitly checked your Smart features setting, Gemini may be reading those communications right now. To verify: go to myaccount.google.com → Data and Privacy → Gemini Apps Activity and confirm the setting status.

4. Agentic Capabilities Expand the Attack Surface (Gemini Spark)

Announced at Google I/O 2026, Gemini Spark is a new agentic mode that can execute tasks autonomously across Google Workspace and, by summer 2026, third-party tools via MCP (Model Context Protocol). As security researchers at Concentric noted in June 2026, this changes the threat model significantly: Gemini is no longer just surfacing information but acting on it. A compensation spreadsheet and a pricing draft look identical to Gemini when both sit in a shared Drive folder — it doesn’t evaluate business context before processing.

5. “Personal Intelligence” Mode: Your Entire Google Footprint as Context

Google has introduced a feature called Personal Intelligence that fuses Gmail, Google Photos, YouTube history, and Search data into a single reasoning context for Gemini. Google’s privacy assurances around this feature leave open questions that security teams at enterprises have begun flagging: where the reasoning happens, how long that fused context is retained, and who has access to the inference infrastructure.

6. No Prorated Refunds on Annual Subscriptions

The Gemini subscription page (confirmed June 2026) states: “No refunds for partial billing periods, except as required by applicable law.” For annual plans specifically, this means if you subscribe at the $199.99/year rate and cancel after three months, you forfeit the remaining nine months of prepayment. Monthly subscribers retain access until the end of the billing period only.

Risks We Did NOT Find

Axis Intelligence does not write hit pieces. The following is what we audited and did not find — and that is part of the record.

• No credential harvesting or phishing infrastructure. Gemini is served entirely from verified Google domains over HTTPS with valid TLS certificates. No suspicious redirects or mixed-content warnings were detected.
• No hidden charges or dark-pattern subscription flows. The upgrade funnel from free to Google AI Pro is clearly labeled, pricing is disclosed before payment, and the subscription confirmation email documents the terms.
• No evidence of selling user data to third parties. Google explicitly states that Gemini conversation data is not sold to advertisers. The company does not use Gemini conversations to serve personalized ads.
• No malware or security vulnerabilities flagged in the app. A review of the Gemini app on Google Play (June 2026) showed no developer warnings, no security disclosures, and a current security update certificate.
• Content moderation is actively maintained. Gemini has hardcoded blocks on content that endangers child safety — these cannot be disabled even via API configuration. Four configurable harm categories (harassment, hate speech, sexually explicit, dangerous content) are also enabled by default.
• Workspace customers are better protected. For enterprise Google Workspace customers, conversation data is classified as customer data under the Cloud Data Processing Addendum and is not used to train public AI models without explicit customer permission. Data residency controls and Customer-Managed Encryption Keys (CMEK) are available in enterprise tiers.
• No age spoofing found to be technically trivial. While age verification is self-reported at signup, Google’s under-13 Family Link controls are functional and actively enforced. Teens 13–17 on Google accounts receive additional content filters, and their conversations are not used to train AI models.

How to Use Google Gemini More Safely

These steps take under five minutes and address the specific risks documented in this audit.

Step 1: Audit your Gemini Apps Activity setting Navigate to myaccount.google.com → Data and Privacy → Gemini Apps Activity. Set auto-delete to 3 months (shortest available). To disable training and human review: turn Gemini Apps Activity off entirely. Note that this limits personalization.

Step 2: Check your Google app permissions on Android Go to Settings → Apps → Google → Permissions. Review each permission individually. Consider revoking access to Contacts, Call logs, and precise Location unless you actively use those features within Gemini or Google Assistant.

Step 3: Disable Gemini Smart Features for Gmail/Chat/Meet if not needed In Gmail, go to Settings → See all settings → General → Smart features and personalization. Toggle off if you don’t want Gemini processing your email content. Repeat in Google Chat and Meet settings.

Step 4: Understand what not to enter into Gemini Per Google’s own guidance, do not enter: passwords or authentication credentials, Social Security or government ID numbers, financial account details, protected health information (PHI), confidential legal or employment information, personal details of others without their consent.

Step 5: For Workspace/Enterprise environments — require CMEK and review access controls Before deploying Gemini to a workforce, ensure your administrator has reviewed the Generative AI in Google Workspace Privacy Hub and configured Data Loss Prevention (DLP) policies. Gemini in Workspace does not retrieve files that IRM policies restrict — but DLP must be configured to enforce this.

Step 6: Review before using Gemini Spark (agentic mode) If your organization is evaluating Gemini Spark, conduct a data access audit before enabling it. Identify all shared Drive folders, and apply appropriate IRM restrictions to sensitive files before Gemini is able to act on them autonomously.

Safer Alternatives

If the data collection scope or the active litigation around Gemini concerns you, these tools offer different privacy postures:

For privacy-first AI chat: → See our guide to best private AI assistants for tools with no-training guarantees and local processing options.

For enterprise AI with stronger compliance controls: → Our review of best AI tools for business covers Microsoft Copilot (Azure OpenAI) and Anthropic Claude, both of which offer enterprise agreements with stricter data handling terms.

For AI in regulated industries (healthcare, legal, finance): → See our breakdown of HIPAA-compliant AI tools — consumer-tier Gemini is not HIPAA-covered without a Business Associate Agreement (BAA), which Google does not offer for Google AI Pro.

Verdict by Use Case

User Profile	Verdict	Reasoning
Casual consumer (searches, creative tasks, general Q&A)	✅ Safe with configuration	Risks are manageable with a 5-minute settings audit. Legitimate platform, strong infrastructure.
Student (13–17)	⚠️ Use with parental guidance	Content filters exist but Common Sense Media rated Gemini “high risk” for minors. No age-differentiated communication by default.
Child under 13 via Family Link	⚠️ Parental supervision required	Google now permits under-13 access through supervised accounts. Filters are active but Google acknowledges they’re not perfect.
Professional handling sensitive data (HR, legal, medical)	⚠️ Caution	Do not input confidential client, patient, or employee data. Consumer Gemini is not HIPAA or attorney-client privilege compliant.
Small-to-mid business user (non-regulated)	⚠️ Caution + configuration	Use Workspace edition with DLP configured, not consumer tier. Review Smart features defaults for all users.
Enterprise / regulated industry	🔴 Not recommended without controls	Requires enterprise Workspace agreement, BAA where applicable, CMEK, DLP, and an IRM review before Gemini Spark is enabled.
Developer using the free API tier	🔴 Not for sensitive data	Google’s API terms explicitly permit use of free-tier inputs for model training with human review. Paid tiers have different terms.

---

FAQ

Is Google Gemini safe to use?

For casual, non-sensitive use — yes. Gemini runs on Google’s enterprise-grade infrastructure, uses HTTPS/TLS on all endpoints, and is a legitimate product of a publicly traded company under active regulatory scrutiny. The risks documented in this audit relate primarily to data collection scope and privacy settings that are not configured by default, not to technical security of the platform itself.

Does Google Gemini sell your data?

No. Google states explicitly that Gemini conversation data is not sold to advertisers or third parties. However, conversations may be used to improve Google’s AI models, and human reviewers may access a subset of conversations for quality purposes. You can limit this by turning off Gemini Apps Activity in your Google account settings.

Can Google employees read my Gemini conversations?

Yes, in limited circumstances. Google employs trained human reviewers (including third-party contractors) who may read, annotate, and process a sample of conversations for safety evaluation and model improvement. Google states it takes steps to disconnect this data from your account identity before reviewer access. Per Google’s privacy documentation, if you delete your Gemini Apps Activity, any copy already accessed by a human reviewer is retained separately for up to three years.

Is Gemini safe for children?

Google requires users to be at least 13 to access standard Gemini. Children under 13 can now access Gemini through Family Link supervised accounts. Content filters are applied to all under-18 accounts. However, Common Sense Media issued a “high risk” rating for Gemini for children and teens in November 2025, citing insufficient age-differentiated communication defaults and the broad data collection scope. Parental supervision and explicit guidance are recommended regardless of the account settings applied.

What is the Thele v. Google lawsuit about?

Thele v. Google LLC (Case No. 5:25-cv-09704, N.D. Cal.) is a class-action complaint filed November 11, 2025. It alleges that on or around October 10, 2025, Google enabled its Gemini “Smart features” — which allow Gemini to read Gmail, Chat, and Meet content — for all accounts by default, without a clear, separate consent prompt. The lawsuit invokes the California Invasion of Privacy Act (CIPA). As of our audit date (June 2026), the case is in discovery. Google denies the core allegations.

Is Gemini HIPAA compliant?

Not in the consumer tiers (Google AI Plus, Pro, Ultra). Google does not offer a Business Associate Agreement (BAA) for consumer Gemini plans. Healthcare organizations subject to HIPAA must use dedicated enterprise solutions and should consult their compliance officer before allowing staff to use consumer-tier Gemini for any work-related task. Google Workspace Enterprise customers should contact Google directly to evaluate BAA availability for their specific configuration.

What happens to my data if I delete my Gemini history?

Deleting your Gemini Apps Activity removes conversations from your visible history and stops new conversations from being stored for personalization. However, per Google’s data retention policy, any conversation that was reviewed by a human quality reviewer is retained separately by Google for up to three years, regardless of your deletion request. Activity across other connected Google services (YouTube, Gmail, Drive) is not deleted by clearing Gemini Activity alone.

Is the Google Gemini app safe to download?

Yes. The Gemini app is distributed through the Google Play Store and Apple App Store with standard security review processes. No malware or unauthorized data exfiltration has been documented in independent security reviews. The privacy risks noted in this audit are product design choices Google has disclosed in its policy documentation — not evidence of malicious software behavior.

---

This audit reflects information available as of June 11, 2026. Google regularly updates Gemini’s privacy practices and product features. Check the Gemini Apps Privacy Hub for the most current policy.