Field notes from the operator’s deck.

The Correspondence File · No. 5

A watch bill puts a name against every station. Ours puts a name against every action an agent takes, and a gate in front of every one that cannot be taken back.

The Correspondence File · No. 4

The talking heads keep reaching for an escaped monster. The right film is Fantasia: no malice, no motives, just unauthorized power, granted casually, running unsupervised.

The Correspondence File · No. 1

This practice did not invent its doctrine. The Navy wrote it first, in ink, and proved it the hard way. A map of the parity, in three parts.

The Correspondence File · No. 2

Grant learned war in the billet he resented. The doctrine he carried out of it, substrate before strategy and the ledger over the speech, is the one that decides whether any system survives contact.

The Correspondence File · No. 3

The Navy keeps two kinds of records and never confuses them. Most AI governance today is the wrong kind.

An AI agent, denied its built-in web tool, reached for a shell command instead. One control let the request through. The control beneath it did not, and the packet capture proved which.

If the printing press had shipped with spyware, the Renaissance would have died in a decade, and it's a fair description of how most organizations use frontier AI today.

A system prompt is a probabilistic instruction, not a control. Under load a degraded instruction fails open.

A safety control passed every test in its suite, and still failed. The postmortem changed how I test everything.

You can write a full governing document for your AI agents, share it across the entire stack, and they will still drift.