On September 2, 2025, Google updated Gemini's terms to give themselves "complete authority to use a sample of users' data to train its large language model." For free users, everything you type into Gemini can now be used to improve Google's AI models, with human reviewers potentially reading, annotating, and processing your inputs and outputs.
Meanwhile, California's Privacy Protection Agency has been busy. CCPA enforcement exceeded $1.3 million in fines in 2025, with joint investigations now targeting businesses across multiple states. On January 1, 2026, new regulations took effect that specifically address AI and automated decision-making technology.
If you're using Gemini with California residents' data, you need to understand exactly where the compliance lines are.
The short version: If you need to redact sensitive documents before they reach AI systems, PaperVeil handles that layer. The rest of this article explains where it fits in the broader governance architecture.
The Direct Answer: Is Gemini CCPA Compliant?
Google states they're committed to CCPA compliance and that they don't "sell" personal information as defined under California law. They provide mechanisms for California residents to exercise their privacy rights through their account settings and data management tools.
But Google's compliance posture doesn't automatically make your use of Gemini compliant. Here's where it gets complicated:
Free Gemini and Google AI Studio: Your inputs can be used for model training and may be reviewed by humans. Even after you delete your activity, Google states that logs can persist for up to 72 hours. When you submit California residents' personal information, you're transferring it to a third party, which triggers CCPA disclosure obligations.
Gemini for Google Workspace (Business/Enterprise): Google confirms that enterprise data is not used for model training and is not reviewed by humans. Customer content stays within the organization and isn't shared externally without permission. These versions bring enterprise-grade security controls and can support regulatory compliance.
Gemini for Google Cloud: Similar to Workspace, prompts and responses aren't used for training. Google has explicitly confirmed this in their 2025 documentation.
The fundamental issue remains the same across all tiers: you're the data controller. When California residents' personal information flows through Gemini, you're responsible for ensuring that processing has proper legal basis and that you've made required disclosures.
What CCPA Actually Requires
The California Consumer Privacy Act applies to for-profit businesses meeting any of these thresholds:
- Annual gross revenue exceeding $26,625,000 (the 2025-2026 inflation-adjusted figure)
- Processing personal information of 100,000 or more California residents or households annually
- Deriving 50% or more of annual revenue from selling or sharing personal information
Meeting any single threshold triggers the full set of CCPA obligations.
Consumer Rights
Right to Know: Consumers can request details about what personal information you collect, where it comes from, why you collect it, and who you share it with. Using Gemini to process customer data is a category of processing you need to disclose.
Right to Delete: Consumers can request deletion of their personal information. If their data has been submitted to Gemini, you need to understand whether and how deletion is possible.
Right to Opt-Out: Consumers can opt out of the "sale" or "sharing" of their personal information. While Google states they don't sell data, if you're using Gemini outputs for behavioral advertising or cross-site tracking, you may be "sharing" under CCPA's definition.
Right to Correct: Consumers can request correction of inaccurate personal information.
Right to Limit Sensitive Personal Information: Consumers can restrict how you use sensitive categories including Social Security numbers, financial account information, precise geolocation, health data, and more.
The 2026 Penalties
CCPA penalties as of January 2025 stand at:
- $2,663 per negligent or unintentional violation
- $7,988 per intentional violation or violations involving minors
These amounts are inflation-adjusted and will be reviewed again in 2027. The per-violation structure means exposure compounds quickly. Processing 10,000 California residents' data improperly could mean $26.6 million to $79.8 million in potential fines.
The 2026 AI Regulations
California finalized new regulations in July 2025 that fundamentally change how CCPA applies to AI. These rules took effect January 1, 2026.
Automated Decision-Making Technology (ADMT)
The regulations define ADMT as any system that processes personal information and uses computation to make or execute decisions or facilitate human decision-making. Gemini clearly falls within this definition when used to inform decisions about consumers.
What's Required
Pre-Use Notices: If you're using ADMT to make "significant decisions" about consumers, you must notify them before processing begins. Significant decisions include employment, housing, financial services, healthcare, and educational opportunities.
Opt-Out Rights: Consumers must be able to opt out of ADMT being used for significant decisions affecting them.
Right to Explanation: You must provide "meaningful information about the logic involved in such decision-making processes, as well as a description of the likely outcome of the process with respect to the consumer."
Risk Assessments: Before using ADMT for significant decisions, processing sensitive personal information, or training AI models on personal data, you must conduct a privacy risk assessment.
Cybersecurity Audits: New audit requirements are phased by revenue, with the first submissions due by April 1, 2028 for companies over $100 million in revenue.
The Explainability Challenge
Try explaining how Gemini's neural network arrived at a specific recommendation about a loan applicant or job candidate. Large language models aren't explainable systems. If you're using Gemini outputs for significant decisions about California consumers, you face a fundamental compliance challenge.
Where Gemini Creates CCPA Risk
Let's be specific about the gaps.
Training on data (free tier): When you use free Gemini, your inputs may be used to train Google's models. A California resident's deletion request doesn't "untrain" the model. You can't fully honor deletion rights for data that's been incorporated into training.
Disclosure requirements: Even with enterprise tiers, using Gemini to process personal information creates a disclosure obligation. Your privacy policy needs to accurately describe this processing activity.
The sensitive data problem: CCPA defines sensitive personal information to include a broad range of data types. As of January 1, 2025, AB 1008 specifies that personal information can exist in AI systems capable of outputting that information. SB 1223 added neural data to the sensitive category. If you're inputting sensitive data into Gemini, you trigger enhanced consent requirements.
Downstream use uncertainty: Once data goes to Google, you have limited visibility into how it's used, even in enterprise tiers. Google's terms allow for "security, monitoring, QA, abuse prevention, and analytics." Residual logs can persist even after deletion.
The Samsung-style risk: In 2023, Samsung employees accidentally uploaded source code and confidential meeting notes to ChatGPT, leading to a company-wide ban. The same risk exists with Gemini. Data containing personal information goes to Google before you realize what was submitted.
The Workaround: Using Gemini While Maintaining Compliance
The solution follows the same pattern that works for all privacy regulations: remove personal information before it reaches the AI.
Document with California residents' personal information
↓
Automated redaction (names, addresses, SSNs, financial accounts, etc.)
↓
Redacted content sent to Gemini
↓
AI processes only anonymized data
↓
Personal information never leaves your control
This approach means:
- No disclosure obligations for Gemini processing (there's no personal information being processed)
- No deletion complexity (the AI never had identifiable data)
- No ADMT transparency issues (decision inputs don't include personal information)
- No training exposure (anonymized data can't identify anyone)
Implementation Steps
Step 1: Map your data flows
Identify everywhere Gemini might encounter personal information:
- Customer service using Gemini in Gmail for response drafting
- Teams using Gemini in Docs for document summarization
- Sales using Gemini to analyze customer communications
- HR using Gemini for employee-related tasks
- Any Workspace integration that touches customer or employee data
Step 2: Classify by CCPA sensitivity
CCPA defines sensitive personal information as including:
- Social Security, driver's license, state ID, or passport numbers
- Account login credentials
- Financial account numbers with access codes
- Precise geolocation
- Racial or ethnic origin
- Religious beliefs
- Union membership
- Contents of mail, email, or text messages (where business isn't the recipient)
- Genetic, biometric, and health information
- Sex life or sexual orientation information
Any document containing these categories requires mandatory redaction before AI processing.
Step 3: Implement automated redaction
Manual review doesn't scale and misses things. Deploy automated detection and removal of:
- Names and identifiers
- Addresses and contact information
- Government-issued IDs
- Financial account numbers
- Health information
- Any California-sensitive categories
Step 4: Establish clear policies
Document and enforce:
- Which Gemini tiers are approved for which use cases
- What data requires redaction before AI processing
- Who can approve exceptions
- How to handle consumer rights requests related to AI processing
Step 5: Update privacy notices
Your California privacy notice needs to disclose AI processing. Be specific about:
- Categories of personal information processed by AI
- Purposes for which AI is used
- Whether ADMT is involved in significant decisions
- How consumers can exercise opt-out rights
Gemini Enterprise Advantages
Gemini for Google Workspace offers stronger compliance positioning:
- Data not used for training by default
- Content stays within organization unless you grant permission
- SOC 1/2/3 certification for operational security
- ISO 27001 and ISO 27701 for security and privacy frameworks
- FedRAMP High certification for government requirements
- HIPAA compliance capability for healthcare data
- Client-side encryption (CSE) for maximum data protection
- Data Loss Prevention (DLP) controls that extend to Gemini
- Data residency options (EU or US)
- Admin controls for centralized user management
Enterprise doesn't eliminate the need for proper data handling, but it reduces baseline risk significantly. For organizations making substantial investments in AI, enterprise tier combined with redaction procedures provides the strongest compliance posture.
The Enforcement Trajectory
California's enforcement is accelerating. The Privacy Protection Agency has made clear that "any violation constitutes a CCPA violation." Using AI without proper safeguards isn't a gray area.
AB 1008's January 2025 amendment specifying that AI systems can contain personal information signals regulatory intent. The state is thinking about AI and building the legal framework to address it.
The 2026 ADMT regulations add another layer. Risk assessments, cybersecurity audits, and explainability requirements are designed to force organizations to think carefully before deploying AI for decisions affecting consumers.
For businesses, the trajectory is clear: more scrutiny on AI, not less. The organizations that avoid enforcement actions will be those that implemented controls before regulators came asking questions.
Your Next Step
The gap between typical Gemini usage and CCPA compliance is real, and California's 2026 AI regulations make it wider. Closing that gap means implementing proper data handling before AI processing, not after.
If you're processing documents that contain California residents' personal information, automated redaction isn't a nice-to-have. It's the mechanism that makes AI usage defensible when the regulator comes calling.
PaperVeil lets you redact all your sensitive information from PDFs in a simple drag and drop flow. Detect and remove PII, match custom patterns, strip metadata, and generate audit trails. The redaction layer that makes AI document processing actually safe.