Trust Center
Every answer is grounded, cited, and refusable.
neolawai answers only from a firm’s own corpus and the legal register. When the evidence is too weak to ground a reliable answer, it refuses instead of guessing — and it verifies every citation before an answer reaches an attorney. These are the measured numbers, not marketing.
Measured groundednessMetrics pending
Loading measured metrics…
Five controls behind every answer
01
Grounded retrieval
Answers draw only from the firm's own precedents and matters plus the legal register — never open-web recall. Retrieval is IDF-weighted so rare, telling terms dominate over boilerplate.
02
Refusal when evidence is thin
A deterministic evidence gate declines before any model runs when retrieval is too weak to ground an answer. Refusal can never itself hallucinate — it works offline.
03
Citation verification
Every emitted [n] reference is checked against the retrieved source range, and each claim sentence is entailment-checked against the source it cites. An answer that cites nothing is treated as a refusal.
04
Immutable audit ledger
Every ask — answered or refused — is recorded to an append-only, trigger-enforced verification ledger: the question, source identifiers, a hash of the checked text, and a timestamp. Immutability is enforced in the database; a write that fails is logged as an audit gap, never silently masked. The trail an attorney shows a carrier.
05
Human in the loop
Output is AI-assisted legal research for attorney review — not legal advice, and never autonomous. A UPL notice rides with every answer, and a confidence score tells the reviewer how far to trust it.
Governance & compliance
Built to survive scrutiny
Credibility is the product. The controls above are the foundation for a formal governance program — here is where each standard stands. We publish status honestly; nothing below is claimed as certified until it is.
Data minimization & encryptionIn progress
In transit and at rest; ledger stores hashes, not client text.
RBAC · MFAIn progress
Tenancy-scoped access with owner/admin roles; step-up (AAL2) enforcement available on privileged admin actions.
NIST AI RMF · ISO 42001In progress
AI governance framing; Measure function is now live via this report.
SOC 2 Type IIPlanned
Controls being formalized ahead of an audit window.
ISO 27001Planned
Information-security management system scoping.
Privacy-law compliancePlanned
Mapping applicable data-privacy obligations.
Want to see the gate refuse, cite, and score in a live matter?
Explore neolawai for firms →Metrics are computed by an automated evaluation harness over a labeled gold set and regenerated on release. AI-assisted legal research for attorney review — not legal advice.