AI Disclosure — Trial Commander
Effective Date: September 1, 2026 (table of AI features updated 17 September 2026, 29 September 2026, 7 October 2026 and 9 October 2026; other AI providers, § 1.1, added 27 September 2026 and updated 29 September 2026 and 9 October 2026; corrected 11 October 2026: Maria, Office and other features added to § 1, §§ 1.2 and 2.3 added, and §§ 1.1, 2.1, 2.2, 4, 6 and 9 corrected to match what the Service does; updated 12 October 2026: §§ 1, 1.1, 1.2, 9 and 11, for the firm-wide AI switch, Practice Mirror becoming opt-in, masking in Office, Maria's notice outside Florida and reporting AI output)
This document describes how Trial Commander uses artificial intelligence (AI) and what this means for your obligations as a Florida attorney.
1. What AI Features Are Included
Trial Commander uses Anthropic's Claude API for these features:
| Feature | What It Does | Data Sent to AI |
|---|---|---|
| Atticus AI | General Q&A about your case | Case context you select |
| Fee Petition Drafting | Generates Rowe lodestar petition | Time entries, case info, your factor analysis |
| Closing Statement | Generates Rule 4-1.5(f)(5) closing statement | Contingent fee calculation data |
| Legal Research | Searches case law with context | Your research query, relevant case background |
| Discovery Responses | Drafts initial responses to discovery requests | Discovery request text, matter facts |
| Cross-Examination Prep | Generates cross-examination outlines | Witness statements, case theory |
| Crucible impeachment triage | Flags deposition testimony that does not line up with the matter's record — live as new answers are captured, or when you run a full scan; each flag names the record it rests on and is shown to the attorney to review | The deponent's name; the session's transcript lines (for live triage, the new answers and the few lines before them; for a full scan, all of them); and entries from the matter's record — key facts, timeline, exhibits, discovery, witnesses, experts, prior depositions and motions — up to a set length (and, once search by meaning is connected, the most relevant passages of the matter's indexed documents). Not sent for a matter set to keep all AI features on the attorney's computer |
| Document Generator | Creates motions, pleadings | Matter facts, legal arguments |
| Ask Trial Commander | Answers a question about using the product, and about your firm's own records, and shows its sources | Your question, passages from the product manuals, and records from the matters you may already open |
| Support replies (Trial Commander staff) | Drafts a reply to a support question from Trial Commander's own help material, only when our staff choose "Draft with AI"; a person at Trial Commander reads, edits and sends every reply | The support question and passages from Trial Commander's own help material; never sent for a firm that requires zero-retention AI or has turned its AI features off (§ 9) |
| Drafting in the Word add-in | Drafts a document from the Drafting Studio into the open document | The matter context you select, your instruction |
| Selection tools in the Word add-in | Explains, simplifies, tightens or rewrites the text you selected | The selected text, the matter it belongs to |
| Cite-check in the Word add-in | Checks the citations in the text you selected | The selected text |
| Reply drafting in the Outlook add-in | Drafts a reply to the open email from the matter | The email you are reading, matter context |
| Firm drafting skills | Derives a reusable drafting skill from the firm's own model documents | The documents the firm submits, and an attorney's approval before use |
| Chronicle time narratives | Drafts a billing narrative for a captured time block, when you ask | The block's times and duration, your notes, the matter number and name, and the titles of the block's sources — including a calendar event's title when you imported the block from a connected Google or Outlook calendar — with Social Security, card and long account numbers masked |
| Maria (client portal assistant) | Answers your client's questions in the client portal about the client's own matter: its status, upcoming deadlines, invoices and trust balance, the documents your firm shared, and how to reach your firm. Replies in English or Spanish. Maria's replies go straight to your client; no attorney reviews them first (§ 1.2) | The client's message and the earlier messages of the conversation, with Social Security, card and long account numbers masked; the matter's name and number, court, court case number, trial date and status; its open deadlines; the invoices the client can see and the total outstanding; the trust balance; the names of the documents, and the channel, subject and date of the communications, your firm shared with the client; the lead attorney's name and email address and your firm's contact details |
| Office (public intake assistant) | On your firm's public intake page, once a firm administrator switches it on: tells the visitor it is an AI assistant, takes down the visitor's name, contact details and a short description of what they need, and passes it to your firm as a lead. Replies in English or Spanish. Its replies go straight to the visitor; no attorney reviews them first (§ 1.2) | What the visitor types and the earlier messages of the conversation, with Social Security, card and long account numbers masked and e-mail addresses and phone numbers replaced by a note that one was given; your firm's name and the states it practices in |
| Judicial twin and opposing-counsel twin | Analyses a judge's, or opposing counsel's, tendencies and style from text you paste in, as internal work product | The text you paste (for a judge, public opinions and orders), up to 30,000 characters, and your question |
| Witness reliability | Reviews a witness's transcript testimony and gives scores, contradictions, gaps and a recommendation on how to use the witness, each tied to transcript citations | The witness's transcript testimony, up to 80,000 characters |
| Client temperature | Only where a firm administrator has switched it on (it is off by default): reports how a client seems to feel about the representation, from the matter's recent messages | Up to 40 of the matter's communications from the last 60 days; filed email only on a matter whose email-AI setting is on |
| Practice Mirror | Only where it is on for your firm: every week, writes each user a private reflection on how they spent their time. Once on, it runs on its own, for each user who logged at least half an hour in the week; only that user sees the report. It is off unless a firm administrator switches it on, except at a firm that already had Practice Mirrors before it became opt-in, where it stays on until a firm administrator switches it off | The user's name; the week's hours by task code, billable and non-billable; practice areas; the names of the matters with the most hours; up to ten of the user's time-entry narratives |
1.1 Other AI Providers
Two features use a provider other than Anthropic, and each runs only once it is set up on the Service (search by meaning, which is not yet connected, is covered below):
| Feature | Provider | Data Sent |
|---|---|---|
| Voice-memo transcription (only once set up on the Service) | OpenAI (model whisper-1) | When set up: the recorded audio of the memo; the transcript comes back |
| Deposition transcription in Crucible (from the effective date of our notice of this change, and only once set up on the Service; never on a matter set to keep all AI features on the attorney's computer, and never while your firm has turned its AI features off, § 9) | Microsoft Azure AI Speech | The recorded deposition audio, with the deponent's and the matter's names as recognition hints; the transcript comes back. The audio is not stored by us |
- OpenAI. Until transcription is set up on the Service, voice memos can be recorded but are not transcribed, and nothing is sent to OpenAI; the voice-memo page says whether it is. OpenAI's published API data controls (read 27 September 2026) state that API data is not used to train its models unless the customer opts in — we have not — and list the transcription endpoint with no abuse-monitoring retention and no application-state retention. Voice-memo transcription is English only for now: every memo is transcribed as English, so a memo spoken in Spanish or another language may come back wrong. Check the transcript against the recording.
- Search by meaning is not yet connected. Search in Trial Commander matches words: the firm-wide search and the Brief Bank find the words you type, and similar-matter matching compares matter fields. No text is sent to an outside provider to be searched by meaning.
- Microsoft Azure AI Speech. Until deposition transcription on our servers is set up, Crucible's Record button uses your browser's speech service (below) or typed entry, and no deposition audio is sent to Microsoft for transcription. Microsoft's published documentation ("Data, privacy, and security for Speech to text", read 9 October 2026) states that for fast transcription, the kind we use, Microsoft does not retain or store the data provided by customers.
- Dictation uses your browser's speech service, not ours. Dictation in Crucible (when no transcription on our servers is set up), matter notes, time-entry narratives, the Ask Trial Commander and Atticus boxes and the staff intake description fields uses the speech-recognition service built into your web browser. Where your browser can transcribe on your own computer, Trial Commander uses that mode and the audio stays on the device. Otherwise your browser may send the audio to its vendor (for example Google, Microsoft or Apple) to transcribe it, under your own agreement with that vendor and not under ours; Trial Commander receives only the text, never the audio, and says so before dictation starts. See Privacy Policy § 4.2.
OpenAI has not signed a Business Associate Agreement with us. Legal Research and citation check use CourtListener (Free Law Project), which is a case-law database, not an AI service. Legal Research sends the words you type into its search to CourtListener; citation check sends only the citation strings, never your draft. The full list is on the Sub-processors page.
Using AI features. You can use Trial Commander without using AI features. Some run without a request from your staff — for example Maria when your client writes to her, Office when a visitor writes to it, and Practice Mirror every week where it is on. A Firm Admin can turn every AI feature off for the firm with one switch; § 9 says what it stops and what it does not.
1.2 Maria and Office: AI That Talks to Your Clients and Prospective Clients
Two features talk directly to people outside your firm.
Maria answers your client in the client portal; she is part of every paid plan. The portal always shows the client a notice that cannot be removed: "Maria is an AI assistant. She can answer general questions about your case status, deadlines, and how to reach your firm. She cannot give legal advice or say how your case will turn out. For legal questions, contact your attorney directly. Conversations are logged per Rule 4-1.6(e)." On a matter that is not a Florida matter, the last sentence reads "Conversations are logged to protect the confidentiality of your information."
- No attorney reviews Maria's replies. Each reply goes to your client as Maria writes it. Maria is instructed to answer only from the matter information listed in § 1, never to invent facts, not to give legal advice or say how a case will turn out, and to send questions about strategy or outcome to the attorney. An instruction is not a guarantee: like any AI, Maria can be wrong. You remain responsible for supervising her use (§ 3.3).
- Your client can reach your firm at any time. One tap sends the client's question to your firm; your staff are notified and the request is kept, encrypted, with the matter. If the AI is unavailable, the client is told so and offered the same button.
- The notice is in English only for now. Maria replies in Spanish when the client has chosen Spanish for the portal or writes in Spanish, but the AI notice above, and Maria's opening greeting, are shown in English.
Office is the intake assistant on your firm's public intake page. It runs only after a firm administrator switches it on (Settings → Office). Its first message tells the visitor that it is an AI assistant, not a lawyer, that nothing it says is legal advice, and that the chat does not create an attorney-client relationship. It is instructed not to give legal advice, not to quote fees, and not to promise confidentiality. Its replies go straight to the visitor; no attorney reviews them first. It replies in English or Spanish, matching the visitor. Before what the visitor types is sent to the AI, Social Security, card and long account numbers are masked and e-mail addresses and phone numbers are replaced by a note that one was given; your firm sees what the visitor actually wrote.
2. How Your Data Is Protected
2.1 Data Retention with Anthropic
Trial Commander sends AI requests to Anthropic's commercial API. Under Anthropic's commercial terms:
- ✅ Your data is used to generate the AI response
- ✅ Anthropic does NOT use your data to train models — in any retention mode
- ✅ Data is encrypted in transit (TLS 1.2+) and during processing
Retention mode. Trial Commander operates in one of two modes, and the active mode is always reflected in the application (Settings → AI monthly limits → Data retention):
- Zero Data Retention (ZDR): once Anthropic activates ZDR on our account, Anthropic retains nothing after the response is generated. We have requested this configuration and will operate under it as soon as Anthropic confirms activation.
- Standard retention (current default until ZDR is active): Anthropic may retain API inputs and outputs for a limited operational period (approximately 30 days) for abuse and safety monitoring, after which they are deleted. They are not used for training and are not accessible for any other purpose.
Firm choices. A firm that requires zero retention before using cloud AI may (a) enable "Require zero retention" in Settings, which pauses the firm's cloud AI until ZDR is active; or (b) connect the firm's own Anthropic API key, in which case requests run under the firm's own agreement with Anthropic. The hosted service does not offer on-device AI.
2.2 In-Product Protections
- AI responses are not shared between firms (each firm's data is isolated)
- Each request to an AI provider is recorded in your firm's audit trail and usage records with metadata only: the feature, the AI model, the size of the request and of the reply (in tokens or, for the streaming assistant, the number of messages and the length of the reply), how the reply ended where the provider reports it, the type and identifier of the record it concerned, who made the request, and when. The text sent to the AI and the text it returned are not written to the audit trail or the usage records.
- Each feature sends the data listed for it in § 1.
2.3 What Trial Commander Keeps
- Atticus conversations on a matter — your questions, Atticus's answers, the actions it proposed and whether they were approved — are kept with the matter. Starting a new conversation closes the old one; it is not deleted.
- Maria conversations — your client's messages and Maria's replies — are kept with the matter in the same way. When the client starts over, the old conversation is closed, not deleted.
- Office conversations are kept with the lead they created (the latest 80 messages of each conversation), for your firm to review.
- What a feature saves — drafts, analyses, reports — is kept where that feature saves it.
All of this is Customer Data. It is kept for as long as your firm's account exists, with no automatic deletion, and is deleted with your firm's other Customer Data after the account is cancelled (Terms of Service § 8). The firm-wide quick question to Atticus and the streaming assistant do not store the conversation; only the metadata above is recorded.
3. Your Ethical Obligations
Using AI does not change your ethical duties as a Florida attorney. Under Florida Bar rules:
3.1 Competence — Rule 4-1.1
You must provide "competent representation." The Florida Bar's November 2023 opinion 24-1 clarifies that:
- You must understand what AI can and cannot do
- You must supervise AI output as you would a junior associate's work
- You must verify citations, quotes, and legal conclusions in AI output before filing or relying on them
- You must not outsource your judgment to AI
3.2 Confidentiality — Rule 4-1.6
Under Rule 4-1.6(e):
- You must take reasonable precautions against unauthorized disclosure of client information
- Using cloud services (including AI) requires informing clients or including cloud use in your engagement letter
- Trial Commander's safeguards (no training on your data, limited or zero retention, encryption) are reasonable precautions, but you remain responsible for disclosing AI use to clients
3.3 Supervision — Rule 4-5.3
AI is a "nonlawyer assistant" under Rule 4-5.3:
- You are responsible for AI output as if produced by a nonlawyer under your supervision
- You cannot delegate attorney judgment to AI
- You must review and adopt AI output as your own work product before filing it
3.4 Candor to Tribunal — Rule 4-3.3
You must not make false statements of law or fact. This includes:
- Not filing AI-generated citations without verification (AI can "hallucinate" citations)
- Not representing AI output as your own independent legal analysis without review
- Disclosing AI use where court rules require it (check your specific court)
3.5 Fees — Rule 4-1.5
You may charge for time reviewing and editing AI output, but:
- You may not charge for the AI's own "work"
- You must be transparent with clients about AI use affecting efficiency
4. Suggested Client Disclosure Language
Consider adding to your fee agreement or engagement letter:
"The Firm may use artificial intelligence (AI) tools, including Trial Commander's integrations with Anthropic's Claude and, where the Firm uses voice-memo transcription, OpenAI, to assist with legal research, document drafting, and case analysis. Client information may be sent to these AI services under commercial terms that prohibit the AI provider from training on or disclosing client data and strictly limit any retention. AI output used in the Firm's legal work is reviewed by an attorney before it is used. [If the Firm offers the client portal's AI assistant: The client portal includes an AI assistant that answers questions about the status of Client's matter directly, without an attorney reviewing each answer; it is not a lawyer and does not give legal advice.] Client consents to this use of AI."
This is a template suggestion. Have counsel review it for your practice.
5. What NOT to Send to AI
As a precaution, avoid sending to AI features:
- Social Security numbers (no valid use case)
- Credit card numbers or bank account numbers
- Medical records containing detailed patient identifiers (use redacted versions)
- Trade secrets or privileged information of third parties (non-clients)
- Content that is subject to protective orders
Trial Commander's AI features don't need this data and you should redact before pasting.
6. What the AI CANNOT Do
AI features:
- Cannot file documents on your behalf
- Cannot electronically sign pleadings
- Are not lawyers. Maria and Office are instructed not to give legal advice and tell the people they talk to that they are AI, not lawyers; but their replies reach those people without attorney review (§ 1.2), so an instruction is not a guarantee
- Cannot access other firms' data, or matters the person using them cannot open
- Cannot access the internet in real-time (except where we explicitly enable it for legal research)
- Remember a conversation only where it is kept: Atticus (on a matter), Maria and Office use the earlier messages of the same conversation, which are kept as § 2.3 describes
7. Malpractice Risk
AI output can contain errors including:
- Hallucinated citations — AI can invent case names, citations, or quote text that doesn't exist. Always verify in Westlaw, Lexis, or free sources.
- Outdated law — AI training data has a cutoff date and may not reflect recent cases or rule changes
- Jurisdiction errors — AI may cite non-Florida law when Florida law applies
- Procedural errors — AI may not know your specific court's local rules
You are responsible for catching these errors before filing. Attorneys have been sanctioned for filing AI-generated briefs with fake citations. See Mata v. Avianca, Inc., 2023 U.S. Dist. LEXIS 107536 (S.D.N.Y. 2023).
8. Verification Checklist
Before filing or sending any AI-generated content:
- I have read the full output
- I have verified every case citation by pulling the actual case
- I have verified quotations are accurate
- I have confirmed cited statutes and rules still exist in their cited form
- The output applies Florida law (not another state's law unless appropriate)
- The output complies with my specific court's local rules and formatting requirements
- I have added my own legal judgment, not just adopted the AI's conclusions
- I have disclosed AI use to my client where appropriate
- I have redacted any info that shouldn't have been in the prompt
9. Turning AI Features Off
A Firm Admin can turn every AI feature off for the whole firm with the switch "AI features for the firm" on Settings → AI Limits. It is on by default, and once turned off it stays off until a Firm Admin turns it back on. While it is off, nothing is sent to an AI model for your firm's AI features, including Maria, Office, Practice Mirror, voice-memo transcription, deposition transcription on our servers in Crucible and AI billed to your firm's own Anthropic API key, and Trial Commander staff answer your firm's support questions without drafting the reply with AI; your client is told only that Maria cannot answer right now and is offered the button that sends the question to your firm (§ 1.2). Everything that is not AI keeps working.
The switch does not stop:
- Dictation, which uses your browser's speech service, not ours (§ 1.1).
- The tools of the AI connector that only read or record your firm's data, used from your own Claude, ChatGPT or other AI (Settings → Connect your AI). Its tools that write with AI stop.
A Firm Admin can also use these controls, on Settings → AI Limits unless stated otherwise:
- Monthly limits. Set a feature's monthly limit to $0 to stop that feature from its next request; set every feature's limit to $0 to stop them all. Limits do not apply to AI billed to your firm's own Anthropic API key, and on some analysis screens a Firm Admin can override the limit for a single request.
- Your own Anthropic API key. If your firm connected one, you can remove it on the same page.
- Require zero retention. Pauses your firm's cloud AI while Anthropic's Zero Data Retention mode is not active. It is a retention control, not an off switch: once that mode is active, the firm's AI runs again.
- Office and client temperature stay off until a Firm Admin switches them on. So does Practice Mirror, except at a firm that already had Practice Mirrors (§ 1); a Firm Admin can switch it off on Settings → AI Limits.
Turning a feature off does not delete anything it already produced.
10. Updates to This Disclosure
As AI technology and Florida Bar guidance evolve, we may update this disclosure. Material changes require 30 days' notice. The current version is always at https://www.trialcommander.com/ai-disclosure.
11. Questions
- General: support@trialcommander.com
- AI-specific: ai@trialcommander.com
- Reporting an AI answer or draft: use Report this output beside an Atticus AI answer or a Drafting Studio draft. Your firm's administrators can see the report under Feedback, and we aim to review it within 2 working days.
- Privacy/security: privacy@trialcommander.com
Florida attorneys remain responsible for compliance with Florida Bar rules and court orders regarding AI disclosure in their specific matters.