AI disclosures

Shield is under active development. Features may change, be limited, or be withdrawn. Free-tier and early-access use carries no uptime guarantee and no support response commitment. Keep your own copies of anything you cannot afford to lose.

Shield routes your requests to AI models and returns generated output. Those models are statistical, not authoritative. Known failure modes include:

• Fabrication — confidently stating facts, citations, case law, standards or clause numbers that do not exist.
• Staleness — answering from training data that predates recent regulatory or product changes.
• Incompleteness — omitting a material exception, condition or requirement.
• Bias — reflecting patterns present in training data.
• Inconsistency — returning different answers to the same question at different times.
• Misreading context — misinterpreting an uploaded document, table or long thread.

Do not use AI output from Shield as the sole basis for a decision in these areas. A qualified human must review and own the decision:

• Legal rights, obligations, filings or regulatory submissions.
• Clinical, diagnostic or treatment decisions, or anything affecting patient safety.
• Employment, promotion, discipline or termination.
• Credit, lending, insurance underwriting or pricing.
• Physical or operational safety, including security incident response.
• Financial reporting, tax positions or audit conclusions.

Shield’s detection, classification, redaction and policy-enforcement features are designed to reduce the risk of sensitive content reaching a model that should not see it. They are automated and imperfect. They may miss sensitive content (a false negative) or block or redact content that was fine (a false positive). Shield reduces exposure; it does not eliminate it, and it does not remove your obligation to control what your people submit.

• Review AI output before acting on it, and verify anything factual, legal, medical, financial or safety-related against an authoritative source.
• Tell the people in your organization that they are interacting with an AI system and that its output requires review.
• Do not submit regulated or high-risk data on a plan that does not cover it — see the restrictions in our Terms of Service.
• Report output that looks harmful, unlawful or badly wrong to shield@dsyfer.com so we can investigate.

Shield is designed to keep a human in the loop: administrators configure which models are permitted, what is redacted, and what is retained, and usage is reported so that AI activity in your organization is visible rather than invisible. Responses generated through Shield are AI-generated and should be identified as such when they are shared or reused.

Requests are served by third-party AI providers acting as subprocessors. Their model behaviour, availability, pricing and acceptable-use rules are outside our control, and zero-data-retention handling depends on the provider and route in use. See the Privacy Policy for how request content and metadata are handled.

Where this site says Shield is “aligned with,” “designed for,” or “supports” a framework such as SOC 2, HIPAA, GDPR or the EU AI Act, we mean that the product is built to help you meet those requirements. Such wording is not a certification, attestation, audit report or legal determination, and it does not by itself make your use of AI compliant. If you need to know which independent reports exist today, ask us at shield@dsyfer.com and we will tell you plainly.

Nothing produced by Shield, and nothing on this site, is legal, medical, financial, tax or security advice, and no professional or fiduciary relationship is created by your use of the Service.