The Tiber River Blog

Notes on private AI for regulated professions

Client confidentiality, on-premises architecture, and the discipline of showing your work. Written by the founder, published when there's something worth saying.

Run a service business? Read the small business blog

November 10, 2026 · Confidentiality

Shadow AI: how to inventory every tool that already touches client files

Phones, browser extensions, and AI features switched on inside software you already pay for. A plain routine for finding every AI tool your staff uses on client work, what to write down, and how to sort it.

Read the post →
November 3, 2026 · Workflow

Scanned documents and honest refusals: why “I can't read this” is the right answer

Faxed records, faded copies, and handwriting get misread, and AI summaries write right over the holes. Why a review aid should name the pages it couldn't read, and how to test a tool with your worst scans.

Read the post →
October 20, 2026 · Positioning

Owning your AI instead of renting it: what changes for a small firm

Most AI tools arrive as subscriptions, which means the vendor can change the model, the terms, and the product without you. What flips when the system belongs to the firm, and the upkeep that comes with it.

Read the post →
October 13, 2026 · Workflow

What a commercial policy renewal review should hand your account team

Material changes in a commercial renewal rarely show up on the declarations page. Why they hide in forms and endorsements, what a cited worklist looks like, and how to test a renewal comparison tool.

Read the post →
October 6, 2026 · Confidentiality

Ethical walls in software: matter-level access control, explained

A screen that lives in a memo depends on everyone remembering it. How matter-level access control works in plain terms, where AI tools tend to punch holes in a wall, and what to test at the demo.

Read the post →
September 29, 2026 · Workflow

What a matter chronology is, and why it eats paralegal hours

Every litigation file ends up needing a fact chronology. Where the hours really go when building one, why a general chatbot doesn't fix it, and what a useful source-linked first draft looks like.

Read the post →
September 22, 2026 · Positioning

The case for boring AI in a professional practice

Demos are built to surprise you. Practices run on work that behaves the same way twice. What predictable software actually means when AI touches client files, and the questions to ask at your next demo.

Read the post →
September 15, 2026 · Governance

What a tamper-evident audit trail is, and why your firm wants one

Every vendor says it keeps logs, and a log isn't an audit trail. What tamper-evident means in plain language, the six things worth recording when AI touches a matter, and five questions to put to any vendor.

Read the post →
September 8, 2026 · Workflow

Why AI output should wait in an approval queue

Your firm already runs on sign-off, and most AI tools give that habit nowhere to live. What a queue entry has to carry, why reviewing in a chat window doesn't hold up, and why a rejection is worth as much as an approval.

Read the post →
September 1, 2026 · Due Diligence

How to read an AI vendor's security page

It's a sales document, and it's still the most useful thing you get before anyone signs an NDA. How to read one for scope instead of adjectives, the three specifics to hunt for, and the four holes that tell you more than the page does.

Read the post →
August 25, 2026 · Architecture

What "your files never leave the building" means, technically

Every AI vendor says something like it now. What a no-egress claim has to cover, the three kinds of data that can slip out of a legal AI system, and the firewall test your own IT person can run in an afternoon.

Read the post →
August 20, 2026 · Confidentiality

Private AI for law firms: why client files should stay in the building

Every AI pitch a law firm hears eventually arrives at the same fine print: send us your documents. The case for refusing, and for running the model on hardware you own instead. Plus five questions to ask any legal AI vendor, including us.

Read the post →
August 20, 2026 · Product Thinking

The verification tax: the real reason lawyers don't trust AI

A confident answer you have to re-check line by line didn't save you time. It moved the work. Why provenance, not intelligence, is the adoption blocker in legal AI, and what output looks like when every line carries its evidence.

Read the post →