AI Output Disclosure

This disclosure describes how the Services use artificial intelligence ("AI") to generate text. It is incorporated into our Terms of Service and complements Section 7 (Data, automated processing, and your mandatory verification obligation).

1. Which models we use

FLREMA's AI text-generation features use large language models ("LLMs") provided by Anthropic, PBC (the Claude family of models). We may change the specific model version from time to time to improve quality or reduce cost. We may also add additional model providers in the future and will update this disclosure when we do.

2. What data is sent to the model provider

When you use an AI tool on the Services, the following may be transmitted to Anthropic for processing:

• The free-text inputs you supply in the tool's form (property descriptions, condition notes, agent observations, buyer/seller context, etc.);
• Property data pre-filled from public county records and other public-data sources (see Data Sources);
• The tool's system prompt (the instructions the model follows);
• Limited metadata necessary for routing the request (e.g., model name, request timestamp).

We do not transmit your password, your DBPR license number, your Social Security or tax ID, your members-only billing information, or other directly-identifying credentials. Your name and email may appear in tool inputs only if you type them there yourself.

3. Anthropic's data handling

Per Anthropic's commercial terms in effect as of our last update, API inputs and outputs are not used to train Anthropic's models by default, and are retained only as needed for operations (typically 30 days for abuse-detection purposes, longer for limited regulatory or safety needs). Anthropic's current policies are at anthropic.com/legal and govern the actual handling. We do not control Anthropic's policies; if Anthropic changes them, the change applies as of Anthropic's effective date.

4. Non-determinism and hallucination

LLMs are non-deterministic: the same input may produce different outputs across runs. LLMs also occasionally hallucinate, meaning they produce text that reads as plausible and confident but is factually wrong, fabricated, internally inconsistent, or omits material context. This is a known characteristic of the technology and is not unique to FLREMA or Anthropic.

We apply guardrails (system prompts that instruct the model not to invent features, not to use Fair Housing protected-class language, etc.) and post-processing sanitization (REALTOR®-mark scrubbing, Fair Housing flag scanning). These guardrails reduce but do not eliminate the risk of incorrect or non-compliant output.

5. Your verification obligation

As required by Section 7 of the Terms of Service, you must independently verify every fact, conclusion, and recommendation in any AI-generated output against authoritative sources before publishing, sharing, or otherwise using it. You must also read every word for Fair Housing compliance, brokerage-policy compliance, and accuracy.

Each AI tool displays an inline verification-acknowledgment prompt that you must check before the output is fully visible and usable. This acknowledgment creates an audit trail of your acceptance of the verification obligation for that specific output.

6. What FLREMA does NOT do with AI output

• FLREMA does not present AI output as fact-checked, attorney-reviewed, broker-approved, or compliant with any specific law, MLS rule, or brokerage policy.
• FLREMA does not make underwriting decisions, pricing recommendations, or legal determinations on your behalf or your clients' behalf based on AI output.
• FLREMA does not represent that AI output is suitable for any particular purpose. Suitability is your judgment to make as the licensed professional.

7. Reporting issues

If you encounter an AI output that is materially wrong, Fair-Housing-noncompliant, or otherwise concerning, please email info@flrema.org with the tool name, the inputs you provided, and a copy of the output. Your reports help us improve guardrails. We do not assume liability for any output you publish or rely on, including output that you previously reported as concerning.

8. Changes to this disclosure

We may update this disclosure as our AI integrations evolve. We will post the updated disclosure here and update the "Last updated" date. Material changes will be communicated to active members.