WASHINGTON DC 3NOV2026
What would an AI that fixes AI have to do? Our new report answers with fourteen tests, drawn from the Vice Chairman’s lecture and from failures running from the Admiralty in 1940 to Minab in 2026. Then it searches the field. Planning tools, kill-chain checkpoints, policy, bias training: each passes a few. One system is built to pass all fourteen. The episode talks through what it is and what it hasn’t proved. Read or listen ↓
The Vice Chairman of the Joint Chiefs warned that speed without judgment just means losing faster. A conversation about the National Security Desk report that says the fix has been built, and has not been proved.
In this episode
• Fourteen tests for any system offered as the fix, and why they come from the Vice Chairman’s lecture and the record rather than from the product
• Why planning AI, kill-chain checkpoints and Directive 3000.09 each meet part of the specification and none governs the judgment chain
• Inside ODAI’s engine room: catalogs of nearly a hundred biases and more than a thousand techniques, five kinds of known, and a flag that cannot be clicked away
• The audit trail as a defense: a record locked in before the outcome is known, and what it would have made of Inchon
• One mechanism for a faster kill chain and for the AI-safety case, after the summer’s laboratory incidents
• What ODAI has not shown, and the shadow test the report says the Vice Chairman can order
Who should read or listen
• Joint planners and targeting staff: where the report says the judgment chain ends and the kill chain begins, and what a flag that must be deliberated would do to a planning cycle
• Requirements and acquisition officials: the proposed wording of a Key Operational Problem, and how the August 2025 reforms, the Joint Acceleration Reserve and SOF RACER could carry an experiment
• Defense AI developers and investors: the report’s test for a new category of AI, and the six gaps it says any competitor must close too
• AI-safety researchers and policy staff: why the report argues the mechanism that makes a force more effective is the one that makes AI safer
• Journalists on the Pentagon and AI beat: the indicators that would show whether the Vice Chairman’s warning becomes a requirement
Why now
The report assesses it likely that the Pentagon’s first fix will go into the kill chain, at nomination and release, and unlikely that the Joint Requirements Oversight Council will rank the judgment chain as a Key Operational Problem in its next annual cycle. The Joint Acceleration Reserve opens with the fiscal 2027 budget; that is the window.
ASSESSMENT HERE
The National Security Desk offers these posts freely, but your support is necessary and appreciated. Please subscribe, paid if you’re able, or leave a tip.
Thank you
