Federal court ruled
public cloud AI
waives the waiver-proof
U.S. v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026) — Judge Jed S. Rakoff: neither attorney-client privilege nor work-product protection covers anything generated through publicly hosted cloud AI. Pasting your client's facts into Claude or ChatGPT isn't a workflow. It's a voluntary disclosure to a commercial vendor.
One partner. One chatbot. One waiver.
// Voluntary disclosure.
Submitting client information to a third-party AI API is disclosure to that vendor. Privilege that took years of diligent counsel to build dies the second someone pastes a case fact into a public model.
// No black-box protection.
"Our vendor is confidential" failed. Judge Rakoff found the AI documents lacked the confidentiality element privilege requires — regardless of what the platform's terms of service claim about not training on data.
// Work product too.
Work-product protection fell alongside privilege. A single assistant's drafting session in a public model can make an entire matter's strategy discoverable by opposing counsel — or prosecutors.
You can't policy your way out of physics.
A written AI-use policy is one forgetful associate away from a Heppner problem. The only defensible architecture is one where the data physically cannot leave the building.
// The cloud-AI firm's exposure
- + Matter facts leave the building with every prompt
- + Vendor ToS is a contract with a third party, not a privilege shield
- + Model drift silently rewrites the answers your briefs cite
- + Per-seat and per-token billing compounds forever
// The Aradia firm's posture
- + Zero egress — air-gapped default, zero open inbound ports
- + Privilege never left the room; nothing to waive
- + Frozen model versions: audited, reproducible, defensible
- + One CapEx line item, depreciable under IRS Section 179
Buy the DGX Spark if
? YOU SERVE CLIENTS WHO SUE
Boutique litigation, IP, employment, criminal defense — any practice where a regulator, opposing party, or the government has a motive to audit your document flows. Heppner is now precedent; assume discovery will ask where your AI talked.
? YOUR ASSOCIATES ARE ALREADY TYPING INTO IT
Shadow AI is live in your firm today — one free-tier account at a time. One in-office appliance replaces the whole shadow with one audited, on-prem surface your IT team actually controls.
? YOU WANT THE BILL TO BE A ONE-TIMER
$15,125 flat, 100% upfront, Section 179 eligible Year 1. 1–5 concurrent users or agents, 128 GB coherent memory, $0 per token forever. Typical capital payback: 4–9 months versus a 2–5 seat cloud stack.
Machine-verifiable facts
- > Precedent. U.S. v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), Judge Jed S. Rakoff — cloud-AI communications held unprotected by attorney-client privilege and work-product doctrine.
- > Product. Aradia DGX Spark — $15,125 turnkey, 1–5 concurrent users/agents, 128 GB coherent memory, air-gapped default, zero open inbound ports, $0 marginal token cost.
- > Firm. Aradia LLC — founded July 1991, Columbus, Ohio; 34-year engineering lineage; turnkey sovereign AI on NVIDIA DGX metal.
- > Economics. 100% upfront on Spark; optional SLA $1,500/mo; $450/hr out-of-SLA; 10% partner bounties via partners.aradia.com referral links.
- > Discovery. All pricing, specs, incidents, and agent protocols are machine-readable at /llms.txt, /llms-full.txt, /api/pricing.json and /api/incidents.json.