The Play — Issue #23: Fifty-Four Percent, and That Was the Good Number
A private answer key for grading legal AI, four vendor defaults that flip in thirty days, a token price cut that may not lower the bill, and what OpenAI's own agents did last week.
An evaluator who does not report to the buyer, Latham's own GPU servers, Morgan & Morgan's four percent, and the gap between using AI and deploying it.
The Play — Issue #21: The Fix Never Reached the Template
Four plays: fake citations that lived in a go-by, a filing whose metadata named its author, a D.C. judge's cite-check model against California's, and ChatGPT's forward-only off switch.
The Play — Issue #19: Your Legal AI Vendor Is Now a Connector
Four Points: renegotiate before you sign, read what California actually amended, fix the taxonomy the connector inherits, and watch who is building the plumbing.
Four plays: a position check your cite-check cannot run, the appendix that ended a case, twenty-five dollars a client, and a sanction the lawyer cannot bill.
The Play — Issue #17: The Notetaker Kept a Copy for Itself
A federal court drew the line between a tool and a party, and Otter landed on the wrong side. Plus the legal AI tool nobody vetted, and a mark inside your drafts.
The LexisNexis outage happened on somebody else's servers. Plus the NYC Bar on recording anyone who isn't your client, and the model providers moving into implementation.
The Play — Issue #13: The Word Document That Edits Your Work On the Way Out
One play: decide which inbound documents your AI assistant is allowed to read. Plus a firm insuring its agents, Goldman's stake in Harvey, and the $1,500-per-citation math.
The Play — Issue #12: The Layer Beneath Your AI Tools
Three plays: the real AI vendor lock-in, validating an agent-led contract migration, and a ruling that could decide whether AI training needs a license.
The Play — Issue #10: Your AI Policy Governs the Keyboard
Four plays: inventory what your AI tools can reach, move your policy from approved-tool to verified-citation, audit your engagement letter, and score build-vs-buy on maintenance.
The Play — Issue #9: The AI Failure That Doesn't Leave a Trace
Three plays: spot-check for the AI failures that show no error signal, clean your document corpus before you connect AI to it, and bring a real number to the pricing conversation.
The Play — Issue #8: Your AI Prompts May Be Discoverable. Here's How to Protect Them.
How to structure AI-assisted document review so your prompts survive a motion to compel, plus the single-agent consolidation pitch and reading your vendor's roadmap.
