Everything a reviewer, regulator, or security team
asks for.

Source Authority

Every answer starts with a question: which sources are even allowed to be used, at what authority
level, under what license, and for what purpose? Source Authority is the control layer that answers this
— before anything is retrieved, not after.

• Authority hierarchy distinguishing primary standards from secondary commentary.
• Version and effective-date tracking so superseded guidance can't silently inform an answer.
• License-aware display, citation, and export rules per source.

AI Safety

Before Kriton™ drafts a response, the request is classified by risk and routed accordingly.

• Low — general concept questions with strong source coverage.
• Medium — workflow guidance requiring explicit assumptions and limitations.
• High — matters requiring stronger source basis and likely human review.
• Restricted— requests outside Kriton™'s intended scope, declined by design.
• Classification Uncertain— routed conservatively when risk can't be confidently assigned.

Platform Limits & Escalation

Kriton™ is designed to clarify, limit, refuse, or escalate rather than answer definitively when source
coverage is weak, a request is restricted, or a matter requires qualified human review. Escalated items
route to a reviewer queue with full context attached — the question, the available sources, and the
reason for escalation.

Data Protection

Tenant data, prompts, uploads, and source bundles are scoped and protected by design — encrypted
in transit and at rest, classified by sensitivity, and subject to retention and regional routing rules before
any model processing occurs.

Access Controls

Access follows a role-based model — preparer, reviewer, admin, and integration service accounts
each receive only the permissions their role requires, configured per tenant.

Responsible AI

Kriton™ is designed to support professional judgment, not replace it. It does not impersonate a licensed
professional, issue binding determinations, certify compliance, or approve filings. When source
coverage is weak, context is missing, or a matter is outside its intended boundary, it's designed to
clarify, limit, or escalate rather than guess.