AI Disclosure

Lane Compliance Systems (“LCS”) is a deterministic-first document pre-review tool. A small number of optional features use artificial intelligence. This page explains, in plain terms, exactly where AI is and is not used, what it can and cannot do, what data those features send and to whom, and the limits you should rely on. It is written to describe our software truthfully and completely. It is not legal advice.

1. The product is deterministic; AI is additive

LCS produces its findings and scores with a deterministic rules engine: required fields, valid date order, consistent identifiers, attached exhibits, and similar checks. These are reproducible and do not involve any AI model. They run entirely in your browser.

On top of that, LCS offers three optional AI features, described below. All are advisory. None of them decides anything. LCS is a first-line, pre-review aid — it is not a decision-maker, not a substitute for professional or legal judgment, and not a lawyer, and it does not provide legal advice.

2. The two optional AI features — what each one actually does

a. AI judgment pass (at review time)

• Off by default. Available only in “Standard” mode and only when an administrator turns it on.
• What it does: adds semantic advisories — for example, whether a narrative actually supports its stated conclusion — alongside the deterministic findings. Each is labeled AI judgment · advisory.
• When it runs: with the pass enabled, it runs automatically right after each deterministic review. Personal identifiers are masked on-device first, and a report that still carries high-sensitivity markers after masking is held back and never sent.
• What it never does: it never changes the deterministic score and never auto-decides. Every AI advisory requires a human to confirm or dismiss it.
• Law-enforcement workspaces: restricted to the organization’s shared key or a self-hosted endpoint the organization controls — the personal in-browser key is not available. Every reviewer is bound to the organization’s choice, and case documents still never go to LCS.
• Where the data goes: the pass runs only if your organization enables it and chooses how to connect its AI. There are three modes, and in none of them does your document text route through LCS: (1) your organization's shared Anthropic key — set once by an administrator and stored encrypted on LCS; at review time it is delivered to the signed-in reviewer's browser so the browser calls Anthropic directly. One key serves every member. Your case documents never reach LCS; the organization's own key does pass through LCS (encrypted at rest, and over HTTPS to your own members). (2) each reviewer's own Anthropic key, entered and held only in that reviewer's browser. (3) a self-hosted endpoint your organization operates — the key never reaches any browser. In every mode the de-identified document text goes only to Anthropic or to your endpoint — never to LCS. We recommend your provider or endpoint contractually guarantee no training on, and no retention of, the data. If you do not want a key delivered to a browser at all, use mode (3).

b. AI-assisted authoring — “Draft with AI” (at setup time)

• Off by default. Helps an administrator turn a free-form writing guide into advisory style rules.
• What it does / never does: it proposes draft rules a human administrator must review and accept; the proposed rules are advisory and never change a finding’s pass/fail call.
• Where the data goes: it sends only the guide/SOP text the administrator typed — a configuration artifact — to the configured AI provider. It never sends the documents you review or any member files. The provider key is encrypted at rest. For drafting it is used server-side only and is never returned to any browser; it is delivered to a browser only if your organization separately enables the shared-key judgment pass described above. An on-device drafting option is also available and sends nothing over the network.

3. What AI in LCS does not do

• It does not make the compliance decision, and does not clear, return, approve, or reject a file.
• It does not change the deterministic score or the deterministic findings.
• It does not run by default — both features are opt-in.
• It does not see the documents you review unless you enable the judgment pass and connect your own AI — either your Anthropic key used directly from your browser, or a self-hosted endpoint you control.
• It does not provide legal advice and does not replace a qualified human reviewer.

c. AI suggested edits (at review time)

• Off by default. When enabled, lets a reviewer ask the AI to propose specific wording changes to the document under review.
• What it does: returns proposed edits — each a specific original span, a suggested replacement, and a short rationale. The reviewer applies or dismisses each one individually. Nothing is applied automatically.
• What it never does: it never edits your document on its own and never changes a score by itself. When you accept an edit, the change is made to your document and the deterministic engine re-scores the new text — so any score change comes only from deterministic re-scoring of wording you approved, never from the AI. A suggestion that cannot be located in your document exactly is shown as advice only and is never applied automatically.
• Where the data goes: identical to the judgment pass above — the same de-identification, the same on-screen confirmation of exactly what would be sent, and the same browser-direct (or self-hosted endpoint) path. Your case documents never route through LCS.

4. Limitations — treat AI output with skepticism

AI advisories and AI-drafted rules can be incomplete, mistaken, or fabricated (“hallucinated”). Treat them as prompts for human attention, not conclusions. A person must verify every AI advisory and approve every AI-proposed rule before relying on it. You remain responsible for the accuracy and the compliance of your work. AI output is not a guarantee of regulatory compliance and is not a legal opinion.

5. AI-generated content is labeled

Inside the app, AI advisories are tagged AI judgment · advisory and AI-proposed rules are shown as advisory drafts that require human confirmation. LCS does not present AI output as a deterministic result or as a final decision.

6. Data handling and third-party providers

• Default — zero document egress. With AI features off (the default), no document or member content is transmitted to any AI service.
• Judgment pass. De-identified document text is sent only to the AI provider or endpoint your organization configures — using your own key directly from your browser, or your self-hosted endpoint — never to LCS.
• Suggested edits. Same path as the judgment pass — de-identified document text goes only to the AI provider or endpoint your organization configures, never to LCS. A proposed edit changes your document only when you accept it.
• Policy grounding (judgment pass). To sharpen the judgment pass, the app retrieves the most relevant passages of your own policy library. Keyword matching runs entirely in your browser with no egress. If your organization saves a Voyage key, the browser additionally embeds a de-identified query directly with Voyage (never through LCS) to rank passages semantically — the same de-identification and Standard-mode gating as the judgment pass itself. Your policy text is indexed with Voyage server-side as a configuration artifact; the documents you review are never sent to Voyage or to LCS.
• Draft with AI. Only administrator-typed guide text is sent to the configured provider (for example, Anthropic). When you enable a third-party provider, your inputs to it are also governed by that provider’s terms — review them, and understand whether the provider retains or trains on inputs.

For the full data map, see our Privacy Policy and Security pages.

7. How we describe our AI

We aim to describe our AI accurately and not to overstate it. In the United States, the Federal Trade Commission enforces against unfair or deceptive practices — including unsubstantiated or exaggerated AI claims and inadequate disclosure of limitations and data practices — under Section 5 of the FTC Act. This page reflects our intent to state our capabilities and limitations plainly and to disclose how the AI features handle data. Our core capability claims are substantiated: the deterministic checks are reproducible and independently testable, and AI is offered only as optional, advisory assistance. This is a plain-language statement of intent, not a representation that any particular legal standard has been met.

8. Not legal advice

This disclosure is provided for transparency and is not legal advice. Organizations in regulated industries should have their own counsel and compliance teams review how they use LCS — including any AI features — against their specific obligations before relying on it. Questions about this disclosure can be directed to your organization’s LCS owner or to the contact listed in our Terms of Service.