● All Verticals — One Surface

Pilot Sturna in your
regulated vertical.

$2,500 deposit. 30-day proof window. $10K citation-error warranty. Five verticals, one standard: no hallucinations reach your regulated workflow.

30-day pilot, deposit credits month 1
$10,000 citation-error warranty
Named compliance contact
HMAC-signed audit trail from day 1
EU AI Act Art. 10–15 conformity doc
See the system running live — Live Metrics →

Five regulated industries.
One hallucination standard.

Each pilot is scoped to your vertical's specific regulatory framework. Same deposit, same 30-day window, same warranty — tuned to the rules you're actually accountable to.

RIA / Investment Advisor
Registered Investment Advisers
Reg S-P (17 C.F.R. § 248) Form ADV SEC 17a-4 Advisers Act § 206
Headline metric: Reg S-P + Form ADV compliance verified on every AI-generated output. Rule 10b-5 anti-fraud screen active.
$2,500 deposit · credits month 1
Medical
Health systems & clinical AI
HIPAA / HITECH 21 CFR Part 11 FDA SaMD EU MDR
Headline metric: PHI interception on every output. SaMD boundary enforcement. BAA-ready from day 1.
$2,500 deposit · credits month 1
Family Office
Multi-generational wealth management
Fiduciary duty UTSA trade secrets 1099-R reporting Dynasty trust
Headline metric: Concentrated holdings analysis + dynasty trust structures verified. 1099-R reporting accuracy enforced.
$2,500 deposit · credits month 1
Actuarial
Insurance & risk modeling
ASOP standards NAIC guidelines ERISA SOA requirements
Headline metric: ASOP-grounded outputs. NAIC reserve calculation verification. ERISA fiduciary screen active.
$2,500 deposit · credits month 1

Triple-Gate proof rail —
every response, every vertical.

Three independent verification layers run in parallel. All three must pass. Any failure is blocked and logged with a tamper-evident HMAC entry.

Gate 1 — Pre-Execution
Intent Classification & Scope Guard
Before any LLM inference runs, the intent is classified against your vertical's regulatory scope. Out-of-domain requests are refused, not guessed at.
Regulatory domain classification
Out-of-scope refusal enforced
PHI / PII pre-scan (medical & family office)
Latency target: <50ms
Gate 2 — Mid-Flight
Citation Grounding & Evidence Check
While the model runs, every factual assertion triggers a parallel citation lookup. No assertion exits unless it traces to a verified source in your vertical's regulatory corpus.
Parallel citation verification
Fabricated source detection
Relevance scoring ≥ 0.70 required
Grounding evidence attached to every response
Gate 3 — Post-Hoc
Competence Screen & Audit Commit
After the response is assembled, a final competence-standard screen checks that conclusions don't exceed the AI's verifiable knowledge. Then the full interaction is committed to the append-only audit log.
Competence-standard enforcement
HMAC-SHA256 chain integrity
Tamper-evident, append-only log
SOC 2 CC7.2 compliant

What you get in 30 days.

Not a generic "pilot." A scoped engagement with measurable outputs tied to your specific regulatory obligations.

Conformity Document
A vertical-tailored EU AI Act conformity assessment covering Articles 9–15: risk management system, data governance, technical documentation, human oversight, and transparency obligations. Ready to present to regulators.
Audit-Log Export
Full 30-day export of your HMAC-SHA256-chained audit trail. Every query, response, citation check, and override gate event. Structured for Drata / Vanta automated compliance platforms.
Workflow Integration
One production integration into your existing workflow — document review, advisory output, clinical query, or reporting pipeline. Grounding and gate enforcement active from the first real transaction.
Named Compliance Contact
A dedicated compliance officer assigned to your pilot — not a support ticket queue. They scope the pilot, set success metrics at kickoff, and present the 30-day review with you.

Not marketing. Proof.

Every claim Sturna makes about hallucination defense is backed by public, reproducible evidence. Check the receipts.

Common questions.