Approved Reference Glossary

A class of machine-learning models trained on massive datasets that produce original content — text, images, code, or analysis — in response to a user prompt. Unlike predictive AI (which forecasts from historical data), generative AI creates new work product. The DRE classifies these tools as "Efficiency Tools" requiring mandatory human oversight. The licensed agent who deploys a generative AI tool remains legally responsible for every output, regardless of which model produced it.

The four-step compliance framework applied at every stage of a California real estate transaction involving AI. Review — read AI output fully and independently verify all factual claims before use. Attribute — record which tool, version, and prompt produced the output, and what review was performed. Correct — treat AI output as a draft; correct errors, remove legally problematic language, apply professional judgment. Explain — be prepared to explain to a client, the DRE, or a court what the AI generated and what you verified. "The AI said so" is not an explanation.

The DRE's foundational requirement that a licensed human professional review and approve AI-generated content before it is used in a real estate transaction or presented to a client or the public. The licensee is the human-in-the-loop. AI tools do not perform licensed acts — licensees do. A licensee who publishes unreviewed AI output has effectively allowed an unlicensed tool to perform a licensed act.

An AI output that is factually incorrect but stated with confidence and internal consistency. Unlike traditional errors — which are typically omissions or clumsy phrasing — AI hallucinations are often specific, plausible, and internally consistent, making them more dangerous because they are less likely to trigger an agent's review instinct. Common in real estate: fabricated square footage, invented HOA rules, non-existent permits, invented school district ratings. The licensee is liable for hallucinations published without verification.

Language in listing copy or marketing materials that signals a neighborhood's racial, ethnic, religious, or familial composition using indirect terms that function as proxies for protected classes. AI models trained on historical real estate data have absorbed decades of such language. Red flags: lifestyle labels ("great family neighborhood"), community vibe descriptors ("quiet, established community"), proximity signals to religious institutions or ethnic business districts, and repeated school district emphasis used as a demographic signal.

The practice of structuring AI prompts to minimize compliance risk. A compliance-first prompt: supplies only verified facts from the TDS and AVID; includes explicit negative constraints on what not to include (neighborhood character, schools, demographic descriptors); assigns a compliance role ("act as a California compliance officer reviewing for Fair Housing violations"); and includes a word count constraint to prevent filler invention. Weak prompts that leave gaps invite AI embellishment and Fair Housing exposure. Prompts must be retained in the transaction file as part of the audit record.

An algorithm-driven property valuation tool that generates a value estimate using comparable sales data, property characteristics, and market trends — without physical inspection. AVMs carry algorithmic bias risk (training data reflects historical market patterns, including those shaped by discriminatory practices) and cannot account for property condition, unreported improvements, or local nuances. AVMs may inform a CMA but cannot replace the licensee's physical inspection, professional judgment, or explanation of basis required under BPC §10176.

The spectrum of AI photo alterations ranked by whether they misrepresent physical property condition. Compliant end: virtual staging (furniture in empty room) and sky replacement. Approaching the disclosure line: temporary object removal. Clearly requiring disclosure: lawn enhancement (dead to lush), exterior crack or stain removal, clutter removal of condition-indicating items. Potentially fraudulent: addition of features that do not exist (pools, decks) or AI removal of known, disclosed defects.

Effective January 1, 2026. Requires that any listing photo digitally altered in a way that materially misrepresents the condition, features, or characteristics of a property carry a clear and conspicuous disclosure that the image has been altered. "Clear" means legible, not buried in fine print. "Conspicuous" means visible in the context where the image is presented. For directly controlled web assets, the original unaltered photo must be displayed alongside the altered image. For print or external platforms, a public link, URL, or QR code to the unaltered original satisfies the rule.

The primary disciplinary statute for California real estate licensees. Key subsections in the AI context: §10176(a) — making any substantial misrepresentation, whether intentional or negligent (covers AI hallucinations published without review); §10176(b) — making any false promise likely to influence a party; §10176(c) — continued course of misrepresentation through advertising; §10176(i) — any conduct constituting fraud or dishonest dealing. A DRE complaint and civil lawsuit may proceed simultaneously. E&O insurance may cover civil liability but does not protect against DRE disciplinary action.

Permits an unlicensed assistant to prepare advertising, documents, and analyses — but only if the final product is reviewed and approved by the licensee before it is presented to a principal or the public. AI is the functional equivalent of an unlicensed assistant under this provision. The licensee's review and approval is the act that converts the AI's preparation into compliant licensed output. Without that review, the agent has effectively allowed an unlicensed tool to perform a licensed act.

Requires a responsible broker to exercise reasonable supervision over all licensed activities of associated salespersons. In the AI context: brokers must establish written office policies governing how licensees use AI tools, ensure AI-assisted outputs are reviewed before use, and maintain oversight of AI tool selection and data handling practices. The duty of supervision is non-delegable — a broker cannot disclaim responsibility for an agent's AI-assisted conduct by pointing to the agent's independent contractor status.

The seller's statutory disclosure of known material facts about a residential property. In the AI context: AI may not prepare or complete the TDS — it contains the seller's sworn representations. When AI summarizes a disclosure package for a client, the licensee remains strictly liable for any missed material facts regardless of whether an AI summary was used. When AI listing copy conflicts with TDS disclosures, the TDS controls and the AI language must be corrected before publication.

The agent's written disclosure of material facts revealed by a reasonably competent visual inspection of the property. Must be completed before AI listing copy is generated — verified facts from the AVID form the only permissible input for AI property descriptions. If an AI draft describes a condition that contradicts an AVID observation, the AI language must be corrected. The AVID, AI prompt, AI draft, and final listing copy must all be retained in the transaction file.

The California Consumer Privacy Act (CCPA) as amended by the California Privacy Rights Act (CPRA) governs how agents handle client personal information when using AI tools. Clients have the right to know what data is collected, the right to deletion (with regulatory exceptions), and the right to opt out of sale or sharing. When an agent inputs client data into a third-party AI platform, the agent is effectively disclosing that data to the platform's retention and sharing practices — which the agent is responsible for understanding.

The regulatory boundary prohibiting non-attorneys from interpreting legal documents, explaining the legal significance of contract terms, or advising clients on legal rights and obligations. AI chatbots and automated transaction agents cross this line when they explain what a contract clause "means for you," advise on remedies, or interpret contingency language beyond factual description. Permissible: summarizing what a document says using the document's own language. Requires referral: anything constituting legal interpretation or advice.

The complete documentation record of AI use in a transaction, required to be retained in the transaction file under BPC §10148. Must include: which AI tool was used; the exact version or model; the prompt as submitted; the AI output as generated (unedited); the final published or submitted version; and a record of what review the licensee performed and what changes were made. The audit trail is the evidentiary foundation for demonstrating that the licensee — not the AI — made the professional judgment.

The fiduciary duty of confidentiality requires agents to protect client information from disclosure to adverse parties or the public. In the AI context, this duty is implicated every time an agent inputs client data — financial information, motivations, negotiation strategy, personal circumstances — into a third-party AI platform. That information becomes subject to the platform's data retention, training use, and third-party sharing policies. Confidentiality risk escalates in negotiation: strategy, offer prices, and client motivation entered into AI tools may be retained by the platform indefinitely.

California requires agency disclosure in residential transactions — clients must know who the agent represents. AI chatbots used for lead generation or client communication must clearly disclose: that the respondent is an AI tool, not a licensed agent; whom the agent or brokerage represents; and that the chatbot cannot provide legal advice or interpret transaction documents. An AI chatbot that allows a prospective client to believe they are receiving representation advice from a licensed agent may violate California's agency disclosure requirements.

An AI-assisted compliance failure can trigger three simultaneous liability tracks: (1) DRE disciplinary action — license suspension or revocation, civil penalty, required education; (2) Civil lawsuit — breach of fiduciary duty, negligent misrepresentation, or fraud claim by the injured party; (3) Fair Housing complaint — administrative complaint to HUD or California Civil Rights Department if the failure involves discriminatory advertising or steering. E&O insurance may cover the civil track but does not protect against DRE discipline or Fair Housing penalties.

The Telephone Consumer Protection Act restricts automated or pre-recorded calls and text messages to consumers without prior express written consent. AI-powered lead generation and automated follow-up systems that send texts or place calls to leads without documented TCPA consent create federal exposure independent of DRE licensing risk. Consent obtained through an AI chatbot must meet the same standard as manually obtained consent. Each unconsented automated text or call can carry a statutory penalty of $500–$1,500.