I write when there is an argument worth making. Planned pieces connect directly to the standards, open problems, and test record.
A concise, downloadable brief on the core proposition: slow the expansion of AI authority without stopping the advancement of AI capability. This companion brief summarizes the current working architecture; it is not presented as the complete position paper.
Download the executive brief (PDF) →A system’s ability to perform an action says nothing by itself about whether it should be permitted to perform it.
Architecture: authority model →Putting a human in the loop is meaningless if the loop does not permit meaningful human judgment.
Architecture: human judgment →Corroboration depends on independent evidentiary roots, not the number of artifacts repeating one upstream claim.
Architecture: evidence independence →Provenance can reconstruct what the system saw while saying nothing about evidence omitted before it arrived.
Open problems →Permission should not persist merely because a technical credential still works.
Architecture: authority model →Passing the original benchmark showed that its cases no longer exposed new weaknesses. So I changed the attack.
Adversarial Lab →These are planned topics, not published articles. I will mark a piece published only when its complete content is available.