Summary

Note: Regulators recognize that “unlearning” from a trained model is technically difficult. Documenting your good-faith efforts and technical constraints is essential for defensibility. - [ ] Determine whether the new use is compatible or requires fresh consent The CCPA requires reasonable security measures. For ML systems, this includes:

CCPA Checklist for Machine Learning: A Complete Compliance Guide

Machine learning systems present unique privacy challenges under the California Consumer Privacy Act (CCPA). Unlike traditional data processing, ML models ingest massive datasets, create derived insights, and can perpetuate privacy risks in ways that aren’t immediately obvious. If your organization trains or deploys machine learning models using personal data from California residents, this checklist will help you build a defensible compliance program.

Why CCPA Compliance Is Especially Complex for ML Systems

The CCPA grants California residents specific rights over their personal information—including the right to know, delete, opt-out of sale, and non-discrimination. Machine learning complicates each of these rights in distinct ways:

Data is embedded in model weights, not just stored in databases
Inferences and derived data may qualify as personal information
Third-party training datasets can introduce hidden compliance gaps
Automated decision-making using ML outputs may require additional disclosures

The California Privacy Rights Act (CPRA), which amended the CCPA, added further requirements around sensitive personal information and automated decision-making that directly affect ML practitioners.

Section 1: Data Inventory and Mapping

Identify All Personal Information Entering Your ML Pipeline

Before you can comply, you need to know what you have. Conduct a thorough data inventory that covers:

Raw training data: customer records, behavioral data, purchase history, location data
Labeled datasets: annotations that may contain personal identifiers
Feature stores: engineered features derived from personal information
Model outputs: predictions, scores, and classifications tied to individuals
Inference logs: records of model predictions made in production

Classify Data by CCPA Category

The CCPA defines specific categories of personal information. Map your ML data to these categories:

Identifiers (name, email, device ID, IP address)
Commercial information (purchase history, browsing behavior)
Biometric information (facial recognition data, voice prints)
Internet or network activity (clickstream data, interaction history)
Geolocation data
Inferences drawn from any of the above

Pro tip: Derived features—even if they don’t look like personal information—may still qualify if they can be linked back to an individual.

Section 2: Privacy Notices and Transparency

Update Your Privacy Policy for ML Use Cases

Your privacy policy must clearly disclose how personal information is used in machine learning. Specifically, you should:

Identify the categories of personal information collected for ML training
Describe the business or commercial purpose of each ML use case
Disclose whether model outputs (inferences) are shared with or sold to third parties
Explain the retention period for training data and model artifacts

Disclose Automated Decision-Making

Under the CPRA, consumers have the right to opt out of automated decision-making that produces legal or similarly significant effects. If your ML model influences credit decisions, employment screening, pricing, or content moderation, you must:

Clearly disclose the use of automated decision-making
Provide a mechanism for consumers to opt out
Offer human review upon request

Section 3: Consumer Rights Fulfillment

Right to Know

When a consumer requests to know what personal information you’ve collected, your response must account for ML-specific data:

[ ] Include data used in model training
[ ] Include inferences generated about the consumer
[ ] Disclose any third parties who received their data as part of model development or deployment
[ ] Respond within 45 days (extendable by another 45 days with notice)

Right to Delete

This is where ML compliance gets technically challenging. Deleting a consumer’s data from a trained model is not straightforward. Your checklist should include:

[ ] Delete the consumer’s raw personal information from source systems
[ ] Remove their records from training datasets
[ ] Assess whether model retraining is necessary (especially if the model was trained on sensitive data)
[ ] Document your deletion process and any technical limitations
[ ] Evaluate whether model outputs (predictions, scores) stored elsewhere must also be deleted

Note: Regulators recognize that “unlearning” from a trained model is technically difficult. Documenting your good-faith efforts and technical constraints is essential for defensibility.

Right to Opt-Out of Sale or Sharing

If your organization shares personal information with third parties for ML purposes—including data brokers, analytics vendors, or model training partners—this may constitute a “sale” or “sharing” under the CCPA. Ensure you:

[ ] Implement a clear “Do Not Sell or Share My Personal Information” link
[ ] Honor opt-out requests within 15 business days
[ ] Audit all third-party ML vendors to confirm they are not reselling data
[ ] Include CCPA-compliant contractual clauses in data processing agreements

Right to Correct

The CPRA added the right to correct inaccurate personal information. For ML systems, this means:

[ ] Provide a mechanism for consumers to flag inaccurate data
[ ] Correct inaccurate records in training datasets where feasible
[ ] Reassess model outputs derived from corrected data

Section 4: Data Minimization and Purpose Limitation

Collect Only What You Need

CCPA and CPRA require that data collection be reasonably necessary and proportionate to the stated purpose. For ML projects:

[ ] Define the specific business purpose before collecting data
[ ] Avoid collecting sensitive personal information unless strictly necessary
[ ] Implement feature selection processes that eliminate unnecessary personal identifiers
[ ] Use anonymization or pseudonymization where technically feasible

Limit Secondary Use of Training Data

Personal information collected for one purpose cannot be used for a materially different purpose without additional disclosure or consent. Before repurposing a dataset for a new ML model:

[ ] Review the original collection purpose disclosed to consumers
[ ] Determine whether the new use is compatible or requires fresh consent
[ ] Update privacy notices if new use cases are introduced

Section 5: Vendor and Third-Party Management

Audit Your ML Vendors

Third-party tools—cloud ML platforms, data labeling services, pre-trained model providers—can create significant CCPA exposure. Your vendor checklist should include:

[ ] Execute Data Processing Agreements (DPAs) with all vendors handling California resident data
[ ] Confirm vendors do not use your data to train their own models without consent
[ ] Verify vendors have their own CCPA-compliant deletion and access processes
[ ] Review vendors’ subprocessor lists for additional exposure

Section 6: Security and Governance

Implement Technical Safeguards

The CCPA requires reasonable security measures. For ML systems, this includes:

[ ] Access controls limiting who can query training datasets
[ ] Audit logs tracking data access throughout the ML pipeline
[ ] Encryption of personal information at rest and in transit
[ ] Model security measures to prevent inference attacks or model inversion

Establish a Privacy Review Process for New ML Projects

Implement a Privacy Impact Assessment (PIA) or Data Protection Impact Assessment (DPIA) process for any new ML project that processes personal information. This should evaluate:

Necessity and proportionality of data use
Consumer rights implications
Risks of re-identification or inference harm
Mitigation measures

Section 7: Documentation and Recordkeeping

Regulators expect evidence of compliance, not just good intentions. Maintain records of:

[ ] Data inventory and mapping documentation
[ ] Consumer rights request logs and response timelines
[ ] Opt-out mechanism testing records
[ ] Vendor contracts and DPAs
[ ] Privacy Impact Assessments for ML projects
[ ] Training records for staff handling consumer requests

Frequently Asked Questions

Does the CCPA apply to inferences generated by machine learning models?

Yes. The CCPA explicitly includes “inferences drawn from” personal information to create a profile about a consumer. If your ML model generates predictions, scores, or classifications tied to California residents, those outputs are personal information subject to the same consumer rights as the underlying data.

What happens if a consumer requests deletion but their data is embedded in a trained model?

This is a recognized technical challenge. The CCPA does not currently require “machine unlearning” in all cases, but you must delete the underlying personal information from your source systems and training datasets. Document your technical limitations honestly, implement model retraining schedules where feasible, and consult legal counsel to establish a defensible policy.

Does using a public or third-party dataset for ML training exempt us from CCPA obligations?

No. If the dataset contains personal information about California residents, your use of that data is subject to CCPA regardless of where you obtained it. You must verify the provenance of any third-party training data and confirm it was collected with appropriate disclosures.

Is a privacy policy update enough to cover ML use cases?

A privacy policy update is necessary but not sufficient. You also need operational processes—consumer request workflows, vendor contracts, deletion procedures, and opt-out mechanisms—that actually deliver the rights you’re disclosing. Policies without operational backing are a compliance risk.

How does the CPRA change CCPA obligations for ML systems?

The CPRA introduced several requirements directly relevant to ML: the right to opt out of automated decision-making, new protections for sensitive personal information (which includes precise geolocation, health data, and biometric data commonly used in ML), data minimization requirements, and the right to correct. Organizations using ML should treat CPRA as the current operative standard.

Take the Guesswork Out of CCPA Compliance

Building a CCPA-compliant machine learning program from scratch is time-consuming and legally complex. Our ready-to-use CCPA compliance template bundle gives you everything you need to move fast without cutting corners:

✅ Pre-built CCPA data inventory and mapping templates
✅ Consumer rights request workflows (know, delete, correct, opt-out)
✅ ML-specific Privacy Impact Assessment template
✅ Vendor Data Processing Agreement (DPA) template
✅ Privacy policy language for ML and automated decision-making disclosures
✅ Staff training checklist and recordkeeping log

Stop building compliance documentation from a blank page. Our templates are attorney-reviewed, CPRA-updated, and ready to customize for your organization in hours—not weeks.

👉 [Browse CCPA Compliance Templates →]