Model Card: FMLA Risk Assessment
This document describes how Sentel uses AI in its FMLA compliance risk assessment tool, what the model can and cannot do, and how results should be interpreted.
What this tool does
The FMLA Risk Assessment analyzes six aspects of your current FMLA compliance process and identifies potential gaps that could create legal exposure. It produces a risk score (0-100), a list of specific findings, and an estimated exposure range per case.
How it works
Your answers to six questions are sent to a large language model (Claude, developed by Anthropic) with a system prompt that constrains the model to act as an FMLA compliance auditor. The prompt contains specific FMLA regulations, scoring calibration rules, and output format requirements.
The model evaluates your answers against federal FMLA requirements, including the DOL Opinion Letter FMLA2025-02-A on shift-worker entitlement calculations, the 7-day cure notice requirement, clock-start rules, and audit trail requirements.
Model details
| Model provider | Anthropic |
| Model | Claude Sonnet 4.6 |
| Prompt version | fmla-audit-v1 |
| Input | 6 structured fields (company size, shift schedule, tracking method, clock start, cure process, recent complaints) |
| Output | Structured JSON: risk level, risk score, summary, 3-5 findings with severity, estimated exposure |
| Human oversight | AI-Assisted, HR-Decided. All results are advisory. No automated compliance decisions are made. |
What this tool is NOT
- ×Not legal advice. This tool does not replace consultation with an employment attorney.
- ×Not a guarantee of compliance. A low risk score does not mean your process is compliant.
- ×Not a substitute for a DOL audit. The tool evaluates self-reported information, not actual documentation.
- ×Not state-specific. The tool evaluates against federal FMLA requirements only. State leave laws may impose additional obligations.
Known limitations
- 1The model relies on self-reported answers. Inaccurate inputs produce inaccurate assessments.
- 2Risk scores are calibrated estimates, not precise measurements. Two companies with the same score may have different actual risk profiles.
- 3The model does not have access to your actual documentation, case files, or employee records.
- 4AI models can produce inconsistent results. The same inputs may produce slightly different scores across runs.
- 5Exposure estimates are based on published case data and statutory ranges. Actual outcomes vary by jurisdiction, judge, and circumstances.
Security and privacy
Your audit responses are sent to Anthropic's API for analysis. Anthropic does not use API inputs for model training. Responses are not stored by Sentel beyond the session unless you create an account. All API calls are logged for security monitoring and quality assurance.
Input validation and injection detection are applied to all fields before they reach the AI model.
How to use the results
Use this report as a starting point for reviewing your FMLA compliance process, not as a final determination. If the report identifies gaps, verify them against your actual documentation and consult with qualified HR professionals or legal counsel before making changes.
For a detailed review of your specific findings by a human, schedule a call.