A lawyer inside technology companies for more than two decades. I write about what AI does to legal work, and this is where I find out whether I'm right. The essays are at kenpriore.com. The papers and the prototypes are here.
A Trust Primitive for Verifiable Autonomous Legal AI Agent Workflows
When an autonomous agent does legal work, nothing today certifies what it did, under what authority, or whether a human reviewed it. The paper specifies that record and maps it onto the EU AI Act and the NIST AI framework.
How Legal AI Agents Can Prove They Were Loyal
Fiduciary duty doesn't disappear when an AI agent makes the decision. It just becomes harder to prove. The paper maps the four fiduciary duties onto the four layers of a sealed record, with tools that exist today.
I took one claim, that the record matters, through peer review, in front of law faculty, and into a prototype that seals what an agent did.
I walk through what makes AI legal work defensible, and you hash the receipt in your own browser.
LiveBoth papers, the ICML poster, and the sealing-pipeline animation live on one page.
LiveThe prototype seals what an agent did across a live pipeline, so you can prove no one tampered with it.
VideoLiveThe model splits the agreement work into three kinds and names who runs each, and it's where to start.
LiveThis part covers the knowledge the agreement work runs on and the record each kind needs.
DraftingMemory is the subject: where it lives, how it reaches the machine, and the three personas the machine works as.
DraftingHere I break a legal call into its steps and mark which ones a machine can learn.
DraftingIt comes down to who assembles the context an agent runs on, and who answers for what's in it.
ResearchingAn agent has to act inside the authority it was given, and someone has to be able to tell when it doesn't.
ResearchingI scored twelve in-house legal roles for AI exposure, and AI takes the process work while the judgment stays with the lawyer.
LiveFive things still protect a company once the drafting gets cheap: trust, context, distribution, taste, and liability.
LiveThis charts where legal-tech funding piled up through Q1 2026.
LiveYou can change one step of an agent's record here and watch the whole chain break.
LiveSoftware engineering moved from writing code to directing it, medicine crossed that line decades ago, and in-house legal is next.
Live