AI & Use of Outputs Disclaimer

340B Audit OS combines deterministic rule engines with AI/ML models. AI is used to: (a) parse uploaded documents into structured fields, (b) narrate gap-analysis findings, (c) draft suggested follow-up questions, and (d) classify and summarize evidence. AI does not decide finding outcomes; severity and pass/fail logic come from the deterministic rule layer.

AI requests are routed through the Lovable AI Gateway (ai.gateway.lovable.dev) to upstream model providers, currently including Google (Gemini family). The set of upstream providers may change over time as model availability changes.

Every AI-assisted output is a forecast or indicator, not a regulatory determination. Outputs may be incomplete, out-of-date, or incorrect. They are not a legal opinion, licensed professional audit, certification of compliance, billing advice, or accounting advice. Do not rely on any output without human review by qualified internal or external professionals.

The Service includes per-finding acknowledgment controls so a human reviewer can mark each automated finding as reviewed before it is relied upon downstream. Use them.

• AI may hallucinate citations, dollar amounts, dates, or document contents that look plausible.
• Document parsing may misread tables, scanned PDFs, or non-standard formats.
• Rule coverage is finite — absence of a finding is not proof of compliance.
• Outputs reflect only the data you provided at the time of analysis.

The Service is designed for non-PHI administrative data only. Customer-side PHI scanners block PHI before it reaches AI providers, but you remain responsible for not submitting PHI (see Terms § 2.3).

Questions about AI usage: [email protected]