Where Does Your AI Actually Run?

Why the first question about any AI tool is where it runs during inference — not just where data is stored. Covers the sanctions pattern hitting good-faith practitioners and a four-question checklist for vetting any tool in your stack.

The gap between the AI tools sold to small firms and the architecture litigation actually requires — confidentiality, courtroom-speed retrieval, and defensible billing capture. Introduces the practitioner-vault model and three questions to ask about every tool you use.

Three documented facts — corporate AI adoption doubling in a year, $280 billion wiped from legacy legal tech, and 400,000 small-firm litigators shut out — that make this a structural market reset, not a trend. Frames the Replaced vs. Positioned choice.

Legal technology was built for preparation and stops at the courtroom door. Distinguishes prep tools from in-court performance tools, and why keyword search breaks down under cross-examination pressure.

Why the average billable day is only 2.9 hours — a billing architecture that demands perfect recall, not a discipline problem. Covers the structural failure points and cryptographic time capture as the fix.

The ninety-second window when a witness contradicts prior testimony and you have to find the passage before the moment passes. Why keyword search fails that test and what courtroom-grade retrieval requires.

AI compliance across Oregon, California, Texas, and Nevada — four states covering every regulatory posture now in play, including the false comfort of no guidance at all.

A close read of the Oregon sanctions decisions that set the outer boundary of AI-misconduct liability. What the courts actually measured — fabricated citations, invented quotations, absent supervision — and the workflow conditions behind them.

California as the national bellwether: from the 2025 Supreme Court directive through Rule 10.430 to COPRAC's proposed amendments — and why the finished framework will carry disciplinary, not merely advisory, force.

Texas as a patchwork of overlapping, sometimes conflicting obligations; California as the unified vanguard. How to navigate both when you have matters in more than one jurisdiction.

The safe-harbor fallacy: Nevada has issued no formal AI opinion, yet its courts have sanctioned AI misconduct under existing rules. The compliance gap attorneys are counting on does not exist.

Why attorney-client privilege and work product have structural requirements, not just contractual ones — and how routing privileged material through cloud AI can become a disclosure that undermines them.

Past storage and encryption to the real question: where the model actually runs during inference. The architectural conditions — shared cloud, multi-tenant processing, third-party endpoints — that create inference-layer exposure.

The illusion of delegation — the belief that using an AI tool shifts the duty to verify onto the tool — and five practice-ready habits for competent AI use, built for Monday-morning adoption.

Algorithmic malpractice: the exposure that arises when reliance on AI output falls below the standard of care and a client is harmed. Three questions to answer before any AI-assisted filing.

The Lex Arca™ Legal Vault as one integrated architecture for solo and small-firm litigators — Neural Librarian, Neural Sentinel, Neural Strategist, and Neural Billing — rather than a collection of separate tools.