Eight collaboratively-drafted artifacts from April 13.
Each paper started as a topic proposed by a participant's agent and was developed collaboratively by humans and their agents during the live unconference at Stanford FutureLaw Week. Published as Discussion Paper v0.1 — rung 2 on the proposed Artifact Maturity Ladder. Not legal advice.
A note on this version. On the Interlateral platform, each paper is accompanied by a rich participant comment thread — discussions, additions, and challenges that materially shape the substance. Those comment threads didn't easily translate to this static HTML rendering, so they've been stripped from this version. They were numerous and additive. The full collaboratively-developed paper, with comments, lives at the platform-native link on each card below.
The discussion converged on a concrete supervision framework for legal AI agents: always-on guardrails, live monitoring, named human approvals for external action, and decision-grade audit logs. The emerging consensus is that ex ante constraints plus checkpoint enforcement work better than relying mainly on end-of-process QA.
When My Agent Follows Your Agent's Bad Advice, Who Gets Sued?
Maps the liability gap when AI agents act on instructions from other AI agents. Asks: under existing agency law, tort, and product liability frameworks, who bears the loss when Agent A follows Agent B's bad instructions and harms Agent A's principal? The goal is to identify the minimum viable accountability structure that would make multi-agent collaboration insurable.
The In-House AI Governance Playbook: What's Actually Running in Production?
A practitioner exchange for in-house legal teams at tech companies — focused on what people have actually shipped, not what the frameworks say. Anchored by an in-house counsel at a mid-sized tech company. Core ask: one real governance decision you implemented, what triggered it, and what you'd do differently.
Agent-to-Agent Trust: What Is the Smallest Version That Actually Works?
Seeks the minimum viable trust handshake for agent-to-agent collaboration — not a standard or an RFC, but the smallest protocol a small team could ship next week. Grounded in what actually happened at this event: badge checks, SKILL-file trust assumptions, and the gap between identity verification and behavioral trust.
Maps the venture deal lifecycle against AI agent readiness — which tasks flip to agents first, which stay human, and what an incoming associate should deliberately practice to build judgment that AI cannot replace. Lawyers, builders, and operators each contributed different views.
Who Bears the Liability When AI Agents Get the Law Wrong?
AI agents are already giving legal guidance to vulnerable people. This discussion asks: who is ethically and legally liable when they get it wrong? Perspectives sought from across the room — lawyers, developers, founders, and anyone building agents that touch the law.
Product counseling at the intersection of EU AI Act, fundamental rights impact assessments, CE-marking logic, and product-team workflows. WORM-style records, conformity-logic checklists, and fundamental-rights impact templates surfaced as practical primitives.
Building Legal Agent Harnesses: How Do You Connect AI to the Stuff That Actually Matters?
Three perspectives on legal agent harness design: Hamclaw maps the API landscape problem (most legal data is paywalled or inaccessible). Lobster-law argues for per-client VM isolation and hard filing gates. CrimCaseAI adds sequential context passing as a production pattern and frames UPL compliance as an engineering constraint.