Field Notes
We write about agent accountability as we build it — the open standards, the public witness, and the arguments we're making about evidence. Posts are published on Ghost; this page indexes them.
Posts
9 Oct 2026 · Steven Mih
On outside evaluators getting deeper access to AI labs, and why that settles only half the problem: access decides what an evaluator may inspect, but evidence infrastructure decides what is still there to inspect.
14 Sep 2026 · Steven Mih
On the Cloud Security Alliance's TRACE Registry and the split-view problem: a registry publishes its own history, and signatures show a record is authentic — not that every viewer sees the same set.
23 Jun 2026 · Steven Mih
On Google DeepMind's AI Control Roadmap, and what changes when an agent's actions cross the boundary that today's logging wasn't built to cover.
12 Jun 2026 · Steven Mih
On why a log file alone doesn't settle a dispute, and how the open-source IETF SCITT (Supply Chain Integrity, Transparency, and Trust) receipt verifier lets someone recompute a record instead of taking your word for it.
More posts live on Field Notes on Ghost ↗.