CCPA / CPRA (California) — Regulatory Reference

CCPA/CPRA requires that businesses honoring consumer opt-out of sale or sharing must propagate those signals to every system that processes that consumer's data — including AI agents. When an agent retrieves personal information to generate a response or recommendation, it is processing data. If the consumer has opted out of sale or sharing, or invoked the right to limit use of sensitive personal information, those restrictions must apply at retrieval time, not just at point of collection. CPRA's Automated Decision-Making Technology (ADMT) rulemaking adds further obligations around transparency and opt-out for consequential automated decisions. Failing to block a retrieval in violation of an opt-out signal is a violation regardless of downstream intent.

CCPA/CPRA applies to for-profit businesses doing business in California that meet one of three thresholds: annual gross revenue above $25M, buying or selling personal information of 100,000+ California consumers or households, or deriving 50%+ of revenue from selling personal information. Technology companies building or operating AI agents that process California consumer data — through product recommendations, personalization, customer support automation, or data analytics pipelines — are within scope if they meet any threshold. The CPRA amendments, enforced by the California Privacy Protection Agency (CPPA), extended obligations to data sharing, not just sale, and added a category of sensitive personal information with heightened restrictions.

CPRA defines a distinct category of Sensitive Personal Information covering Social Security numbers, financial account details, precise geolocation, health and biometric data, race or ethnic origin, religious beliefs, union membership, and contents of private communications. Consumers have the right to direct businesses to limit the use and disclosure of their SPI to what is necessary to provide requested services. For AI workflows, this means an agent must not retrieve SPI for purposes beyond the transaction the consumer initiated. AutoPIL enforces this as a sensitivity ceiling: sources tagged as sensitive personal information under policy TEC-CCPA-SPI-001 are blocked from agent retrieval when the purpose is outside the permitted scope, before the data enters the agent's context window.

AutoPIL addresses two specific CCPA/CPRA obligations. First, opt-out propagation: when a consumer exercises the right to opt out of sale or sharing, that signal must reach every processing system. AutoPIL policy TEC-CCPA-OPT-001 gates agent retrievals against opt-out status — an agent cannot retrieve a flagged consumer's data for a restricted purpose regardless of which framework or pipeline is making the call. Second, the tamper-evident audit log produces a verifiable record that a retrieval was blocked, supporting Right-to-Know responses and demonstrating compliance during a CPPA investigation or audit. Every enforcement decision — allow or deny — is recorded with a cryptographic chain hash that cannot be altered retroactively.

The California Privacy Protection Agency can impose civil penalties of up to $2,500 per unintentional violation and $7,500 per intentional violation. There is no cap per investigation, meaning a systematic failure — such as an AI agent ignoring opt-out signals across thousands of consumer records — can compound to material exposure quickly. The CPPA has authority to audit businesses proactively, not only in response to complaints. For technology companies, the ADMT rulemaking adds potential obligations to provide opt-out mechanisms for automated decisions with significant effects. The CPPA has signaled enforcement focus on data minimization and purpose limitation failures, both of which are directly implicated by AI agent deployments that retrieve more data than the transaction requires.