Submission Desk · Legal
AI Use & Transparency
Contents
1.How AI is used
AI assists with two jobs in Submission Desk:
- Extraction: reading your uploaded documents and producing a structured risk summary (insured details, exposures, locations and values, loss history, limits requested, and a list of what's missing). Each fact is labeled with the document it came from, or "Broker notes".
- Drafting: writing a tailored email to each underwriter, modeled on samples of your own writing, and revising it when you ask.
AI never sends anything, never contacts underwriters, and never makes decisions about coverage, pricing, or eligibility. It produces drafts for you to review.
2.Two ways to use AI
| Option | How it works | Whose terms apply |
|---|---|---|
| Claude connector (primary) | You add Submission Desk to your own Claude account as a custom connector. Claude reads your submission data through the connector when you ask it to, and saves summaries and drafts back. | Your agreement with Anthropic and your Claude plan's data settings. Anthropic is your vendor, not our subprocessor, for this path. |
| Your own Anthropic API key (optional) | You enter a key from the Anthropic Console. Our servers send document text to Anthropic's API using your key. | Anthropic's commercial terms under your API account. No training on your data by default. |
| Our API key (optional, when enabled) | Our servers send document text to Anthropic's API using our account. | Anthropic's commercial terms under our account; Anthropic acts as our subprocessor. No training on your data by default. |
With the connector, what Anthropic does with the data (including retention and whether it may be used to improve models) depends on your Claude plan and settings. Review them in your Claude account, particularly if you use a consumer plan.
3.Grounding rules
The Service instructs the AI to work only from:
- your uploaded documents and the risk summary you reviewed;
- notes you typed;
- your writing samples (for tone and style only, not facts); and
- the appetite notes you saved for that underwriter.
It is told never to use its general knowledge about the insured or the carrier, never to fill a gap with a plausible guess, and to report missing information as missing. AI output is checked against a strict format before it is shown to you, and extraction never silently overwrites your edits: a new run is shown as a proposal you accept field by field.
These safeguards reduce errors. They do not eliminate them. Drafts are written to sound like you, and outbound email does not mention AI.
4.Human review is required
You are the last check
Before sending, compare every figure, limit, loss, value, date, and statement in the draft with the source documents. You send each email yourself from Outlook, and you are responsible for its content, as the Terms explain. There is no automatic-send option, by design.
5.No training on your data
[Company Legal Name, LLC] does not use your documents, drafts, or client information to train or fine-tune AI models, and does not permit our subprocessors to do so. We do record the number of AI tokens used per account, for cost tracking and plan limits.
6.Known limitations
- AI can misread scanned pages, handwriting, rotated or low-quality images, and complex tables.
- It can attribute a figure to the wrong location, year, or line of coverage.
- Very large packages are trimmed to a text budget before being sent to the AI; when that happens the Service tells you, and details in the trimmed portion may be missed.
- It may state things with more confidence than the source supports, or miss something that is present.
- Style imitation is approximate; review tone as well as facts.
7.Using the Service without AI
AI is optional. If you don't connect Claude and don't add an API key (and we haven't enabled in-app AI for your account), none of your data is sent to an AI provider. You can still upload documents, enter and edit the risk summary yourself, write emails using templates, create Outlook drafts, and track results. You can disconnect Claude or remove your key at any time from Connections.
8.Questions and feedback
If AI output was wrong in a way that worries you, please tell us at [legal@yourdomain.com]. Don't include client data in the email unless we ask for it through a secure channel.