Docs & Thoughts

Two federal courts, one day, opposite conclusions on whether chats with generative AI are privileged. Notes on the February 10, 2026 Heppner and Warner rulings.

The disruption thesis was that AI would let small builders out-compete incumbents. Project Glasswing, which grants fifty-two institutions privileged access to an unreleased frontier model, shows the opposite: architecture decides winners. Frontier capability plus centralized billing plus partner-gated distribution equals incumbents winning by default and calling it security policy.

How a mocked hypothetical in the birthright citizenship case quietly maps the constitutional basis for global prosecutorial reach. Justice Jackson's thought experiment at oral argument in Trump v. Barbara embeds a framework where citizenship confers allegiance, allegiance determines prosecutorial reach, and defining who owes allegiance defines who can be prosecuted, and where.

For CTOs, CISOs, and security auditors weighing a third option between adopting a new AI vendor and banning AI outright. Covers how the library brings Shadow AI into compliance by routing it through already-vetted organizational subscriptions, the three deployment architectures (Direct Subscription Utilization, Zero-Interception Routing, Zero-Knowledge Orchestrator), and why none of them creates a new data processor.

Why professional-grade legal automation respects the ethical boundaries of the profession, and serves its deeper purpose.

A local library that routes AI inference through an organization's existing subscriptions. No new data processor, no shadow AI, and the relay never sees a developer's system prompts, tool schemas, or proprietary logic. Secret sauce and data alike are protected by structural impossibility, not policy.