A founder's letter.

The first billion-dollar AI verdict has not happened yet. It will. When it does — when an autonomous agent wires money to the wrong address, denies a claim incorrectly, or executes a contract no human reviewed — the question in the courtroom will not be whether the AI did it. The question will be what evidence exists to prove what it did. And whether that evidence holds up under cross-examination from opposing counsel.

Today, that evidence is a LangSmith trace. A Datadog dashboard. A Sentry log. None of those survive a Daubert challenge. None of those satisfy FRE 902(14) self-authentication. None of those clear an underwriter's chain-of-custody review. They are debugging artifacts produced for engineers — not legal evidence produced for the adversary.

Verdict exists because the evidence layer was missing. Three regulatory deadlines — Verisk AI exclusions in January 2026, the Gartner General Counsel mandate in April, and the EU Product Liability Directive in December — are converging into a forced-buy window. Every enterprise with autonomous agents in production needs evidence that holds. Most do not have it. None of the existing observability or governance vendors are building it, because building it requires cryptographic infrastructure they don't have, legal expertise they don't have, and an insurance partnership posture they haven't taken.

We took it. Three USPTO patents filed defensively. The standard published Apache 2.0 with a perpetual royalty-free license. Reference integrations with three AI liability carriers. A founding use case that started with a $340K wire transfer no one could prove was authorized.

The seal is the evidence. The standard stays open. The category gets locked in 18 months — by us, or by someone else. We're betting it's us.

— Shayne