AI Output and Verification Disclaimer

1. How Vouchstone Reduces Error

Vouchstone is designed to make AI outputs more reliable and auditable. It extracts facts through a multi-pass pipeline, scores their confidence, checks them against formal logic and constraints, holds low-confidence facts for human approval, answers questions through a council of AI advisors that cite their sources, and records decisions in a signed, hash-chained audit ledger.

These features reduce risk; they do not eliminate it. Terms such as "deterministic," "verified," "hallucination-resistant" and "mathematically verified" describe our engineering approach and the consistency of our processing. They are not a guarantee that any fact, answer, mapping or document is correct, complete or fit for your purpose.

2. Limits You Should Understand

• Garbage in, garbage out. Verification checks consistency with the data and rules you provide. If source data, ontologies or business rules are wrong or incomplete, Output may be wrong while appearing verified.
• Model limitations. Large language models can misread documents, omit information or produce plausible but incorrect statements, even with multiple checks.
• Confidence scores are estimates. A high score does not mean a fact is true; thresholds you set affect what is promoted automatically.
• Agents act on instructions. An agent within its permissions can take actions that are technically permitted but commercially or legally wrong for your situation.
• Deterministic documents. emboss-pdf produces the same file from the same structured input. It does not validate that the input data is accurate.

3. Not Professional Advice

Output, including contract reviews, legal research, clinical summaries, financial models, compliance reports and migration mappings, is not legal, medical, financial, tax, accounting, engineering or other professional advice. It does not create a professional-client relationship. Qualified professionals should review Output before it is relied upon for decisions in their field.

4. Your Responsibilities

• Review and validate Output in proportion to the risk of the decision it supports.
• Configure autonomy levels and human-approval gates appropriate to each agent's task.
• Test migrations, mappings and automated workflows in non-production environments before production use, and keep backups.
• Keep humans in control of decisions with legal or similarly significant effects on individuals.
• Identify and meet any legal or regulatory obligations that apply to your use of AI, including human oversight, monitoring and informing affected people. Vouchstone's oversight, logging and transparency features are security and governance tools; they do not discharge those obligations for you.

5. No Compliance Representations

Vouchstone focuses on security and governance controls: scoped access, human-approval gates, verification and signed audit logs. Vouchstone does not represent or certify that the Services, or your use of them, comply with any law, regulation, standard or industry framework. You are solely responsible for determining whether the Services are appropriate for your regulatory obligations and for meeting them.

6. Liability

Your use of and reliance on Output is at your own risk and subject to the warranty disclaimers and limitations of liability in the Terms of Service or your applicable agreement. Questions: legal@vouchstone.ai.

Vouchstone LLC, 2000 Walnut Ave, Fremont, California 94538, United States