Every AI surface in OllieSafe runs behind the same governance frame — jurisdiction resolution, verified citations, confidence tiers, and human control. Chat, photo analysis, voice capture, and drafting all answer to the same rules.
Confidence tiers on every AI feature class
INFORMATIONAL · ADVISORY · DETERMINATIVE
Cases in the governance evaluation
Run before any model change ships
AI feature classes evaluated
Chat, vision, voice, and drafting included
Chat, photo analysis, voice capture, and drafting all run behind the same four layers. None of them are optional, and none of them are policy PDFs — they are how the product works.
Before the AI answers, it resolves the jurisdiction of the physical work — the incident case first, then the establishment, then tenant headquarters, then the federal fallback. Where a jurisdiction has no authored vocabulary yet, the AI answers from the federal floor and discloses the fallback.
Proposed regulatory citations are checked against retrieved regulatory text before they persist. A citation that cannot be verified is flagged in the artifact, and ungrounded identifiers are structurally blocked from becoming part of your record.
Every AI feature class carries one of three confidence tiers — INFORMATIONAL, ADVISORY, or DETERMINATIVE. Determinative outputs route to a human review queue before anything acts on them, and every reviewer decision is audit-logged.
Advisory and determinative outputs land in a hash-chained AI decision log — append-only and tamper-evident. Model changes ship through a 1,435-case governance evaluation across 37 AI feature classes.
Guardrails only count when they are structural. These are enforced in the product, not promised in a policy.
The AI prepares drafts and determinations. Acting on them — including any transmission — remains your explicit act.
A citation that cannot be verified against retrieved regulatory text is flagged in the artifact — visibly, not in a footnote.
Determinative outputs wait in the review queue until a person approves or rejects them — and that decision is audit-logged.
When authored jurisdiction content is not available, the AI answers from the federal floor and discloses the fallback.
Create your account, pick a plan, and start building your safety system of record across all 51 OSHA jurisdictions.