AI Transparency Notice.

Lexboard uses AI to speed up the heavy mechanical parts of personal-injury practice — summarizing medical records, drafting first-pass demand letters, scoring intake leads, chatting with deposition transcripts — but every AI output is a draft requiring attorney review before it touches a client, a court, or a third party. We do not automate legal judgment. We do not let AI send anything outside the firm without explicit human approval.

• AI-Guided Intake — a conversational interview that captures lead facts, scores viability, and writes the AI summary on the lead record.
• AI Case Analysis — strength, key risks, missing-information assessments on demand for any case.
• Medical Intel / Chronology — extracts dates, providers, diagnoses, and treatments from uploaded medical records into a sortable chronology.
• Demand-Letter Drafting — given case facts and jurisdiction, generates a first-pass demand under the appropriate framework (Stowers, Holt, FL unconditional tender, etc.).
• Deposition Chat — semantic search and Q&A across uploaded deposition transcripts.
• Document Template Fill — populates firm templates with case data; the AI is used only when explicit free-text generation is required.
• Trial Prep Co-Counsel — brainstorming partner for opening / closing / cross outlines.

Each AI feature has a visible “AI” badge in the UI and a disclaimer confirming the output is assistive only.

Lexboard routes AI requests to one of two providers based on the feature and per-firm configuration:

• Anthropic — Claude models, via the official Anthropic API.
• OpenAI — GPT models, via the official OpenAI API.

We use only the official enterprise API tiers from these providers — never consumer ChatGPT or Claude.ai web UIs. Under their enterprise terms, neither provider trains their foundation models on data submitted via the API.

When you invoke an AI feature, Lexboard sends to the chosen provider only the data needed to fulfill the request:

• Case facts visible in the surface you triggered (jurisdiction, accident description, injury notes, damages summary).
• For medical-intel features, the uploaded medical-record text.
• For demand-letter drafts, the structured variables of the chosen template.
• System prompts that define the AI's role and constraints.

We do notsend your firm name, your colleagues' names, your authentication tokens, your billing records, or data from other cases the AI request isn't scoped to.

Where the feature's purpose allows, we redact personally identifying details (SSN, DOB, financial-account numbers) before transmission. End-clients' first names and case narratives are passed in their original form because the drafted output needs them.

• Processes the prompt and returns a response.
• Logs the request for short-term abuse monitoring (typically 30 days).
• Does not use the data to train foundation models (per their enterprise terms).
• Does not share the data with other customers.
• Does not sell or resell the data.

See the providers' data-usage statements: Anthropic Commercial Terms, OpenAI Business Terms.

The AI features are not:

• A licensed attorney.
• A substitute for attorney review of any document or communication that will be filed, sent, or relied upon for legal advice.
• A binding statement of the law or of any client's rights.
• A privileged communication channel — sending case data to the AI is processing by a third party (the AI provider) subject to the firm's ethical analysis under Rule 5.3.
• An expert witness or a source admissible in court.

Large language models can produce plausible-sounding text that is factually wrong (“hallucinations”). This is a known limitation of the technology. Lexboard mitigates this by:

• Grounding AI features in the firm's own structured case data wherever feasible.
• Requiring attorney review before any AI output is sent externally.
• Surfacing citations and source paragraphs in deposition-chat and medical-intel features.

These mitigations reduce, but do not eliminate, hallucination risk. The firm bears the final responsibility for verifying every AI output before action.

Firm administrators may, in account settings:

• Disable AI features entirely for the firm.
• Choose which AI provider (Anthropic / OpenAI) handles each feature.
• Bring their own API key (BYOK) so AI calls bill against the firm's own provider account.
• Set monthly cost caps and per-user spend limits.

Every AI invocation is logged in ai_usage_logwith the user id, case id (where applicable), feature, token counts, and cost. Firm administrators may export this log at any time. Output is not silently retained beyond the original case's audit trail.

Lexboard's AI features are designed to support rather than substitute for the licensed attorney's judgment — consistent with ABA Model Rule 1.1 Comment 8 (technological competence) and Rule 5.3 (responsibilities for nonlawyer assistance). See /legal/bar-rules for the full mapping of platform behavior to each rule.

When we add a new AI feature, switch providers, or change what data is transmitted, this page will be updated and firm administrators will be notified by email.

Questions about AI usage or to request a model-specific compliance brief: legal@lexboard.net.