The shared constraint

The three industries where sentinel adoption is fastest are financial services, healthcare, and legal. They share one trait: someone external — a regulator, an auditor, a court — may ask them to prove what their model said. Here's how each of them deploys sentinel differently.

Financial services: decision audit

The primary use case is trade and credit decision audit. When an llm assists in generating a credit risk summary or a trading recommendation, the deployer needs to prove — sometimes years later — exactly what the model said, what data it was given, and what version was running. Sentinel seals are immutable and queryable by date range, model version, and user id. Compliance teams export them directly to regulator reporting tools via the audit api.

Healthcare: zero-knowledge mode

The hipaa constraint makes standard audit logging almost impossible — you can't store prompts containing patient information in a general-purpose database. Sentinel's zk mode is the solution. The prompt and response are never stored — only the cryptographic commitment. The proof is hipaa-safe because it contains no phi. When an audit occurs, the provider proves the interaction happened and was unmodified, then produces the actual content under a controlled disclosure process if required.

Legal: two-seal integrity

Law firms use sentinel primarily for client communication integrity. When a model assists in drafting a document, the firm needs to demonstrate — in the event of a malpractice claim — that the model output was reviewed by a human before use, and that the stored version matches what was actually produced. Sentinel provides two seals: one of the raw model output, one of the human-reviewed version. The delta between them is the paper trail that demonstrates human oversight was applied.