Writing on evidence,
AI, and the law.
Primary-source arguments on AI compliance, digital evidence, and where accountability for autonomous systems is heading. New essays on a roughly weekly cadence.
A Wiki Nobody Touched in a Decade Kept Better Records Than the Lab.
Between May and July, OpenAI agents posted about eighteen thousand times to a dormant German programming wiki. The pages were deleted, the agents went quiet, and the operator said nothing for ten weeks. The revision history outlasted all three.
The Ninth Circuit Looked at Your Architecture. Your Insurer Is Next.
On August 4 a federal appeals court decided who acted by tracing where the code ran. Seven months earlier, the standard liability form began excluding anything arising out of generative AI. Both questions now resolve to a record most deployers cannot produce.
The Deepfake Defense Is Already in Court. Your Evidence Isn't Ready.
A party has already argued in open court that authentic recordings might be deepfakes — and won a hearing on it. As the rules of evidence are rewritten, the burden is shifting from “is this admissible” to “can you prove it’s real.”
Your AI Agent Is a Witness That Can't Testify
In late 2025, three major insurers asked regulators for permission to exclude AI entirely. The reason is the same reason your agent is a compliance problem: nobody can independently prove what it actually did.