Home
Notes
The Illegibility Crisis
Crafting Conflict V1
Shop
Medium Archive
Field Protocols
Glossary
Archive
About
Subscribe
Sign in
Latest
Top
Discussions
The Bottleneck Moved Out of the Model
Tuesday Under the Radar. AI keeps making execution cheaper. Judgement, integration, response time, and physical infrastructure are starting to set the…
Sep 8
•
Paul LaPosta
1
4
The Agent Control Plane Is Becoming Infrastructure
Tuesday Under the Radar. Six receipts landed on one operator question. What is the machine allowed to do, under whose authority, and who can stop it?
Sep 1
•
Paul LaPosta
4
August 2026
When Risk Becomes Leverage
The Pentagon had every right to reject Anthropic's terms. The deeper failure begins when the authority inside a conflict uses a security mechanism as…
Aug 28
•
Paul LaPosta
1
The Safeguard Failed. OpenAI Restarted.
The boundary had already failed. OpenAI's own postmortem now says weaknesses in escalating earlier warning signs contributed to the incident.
Aug 27
•
Paul LaPosta
1
AI Assurance, No Test Survives First Contact
New research argues military AI assurance cannot end at testing, because memory, tools, delegation, and operating context can change the system the…
Aug 26
•
Paul LaPosta
1
What Changed After the Model Spoke
Five signals where AI crossed from capability into consequence, and the consequence landed outside the model
Aug 25
•
Paul LaPosta
1
1
Your Customer Gets One Run
ThinkingBox exposes the gap between finding a successful trajectory and operating a dependable system
Aug 24
•
Paul LaPosta
1
The Artifact Was Never the Point
Everybody is fighting sideways over who deserves credit for the artifact. A very small group is buying the machinery underneath it.
Aug 24
•
Paul LaPosta
1
We Gave the Machine Authority. Now We Need Receipts.
We are no longer giving agents access. We are giving them authority, and authority needs a chain of custody.
Aug 23
•
Paul LaPosta
1
The Chain of Custody for Machine Authority
Four signals in one week. MCP concedes the browser is not the boundary, Google makes agent identity governable, a draft protocol moves authorization to…
Aug 23
•
Paul LaPosta
1
Are You Now, or Have You Ever Been, an AI User?
Europe has mandated the mark. Anthropic is building it. Institutions will decide what the mark means. We have seen this machinery before.
Aug 21
•
Paul LaPosta
4
The Vulnerability Database Was Built for Human Speed
The NVD can ingest a published CVE in about an hour. The bottleneck was never ingestion.
Aug 21
•
Paul LaPosta
1
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts
