Insights

Research notes on AI security, extension risk, and offensive testing.

Tax the chips, before the models
·Ali Mosajjal

Tax the chips, before the models

Keynes promised his grandchildren a fifteen-hour week. The wealth arrived; the leisure didn't. A proposal to tax AI at the hardware layer, per chip, and bank the surplus as time.

Read
Harness the Power of the Harness
·Ali Mosajjal

Harness the Power of the Harness

Google Big Sleep, Anthropic Glasswing, Microsoft MDASH, evilsocket's audit, Hadrian OpenHack, OpenAI Aardvark. Different teams, one recipe. The security-agent harness has become the product, and the model behind it almost doesn't matter.

Read
Hours, Not Days
·Ali Mosajjal

Hours, Not Days

AI-assisted exploit development is collapsing patch timelines. The answer is not louder change management; it is architecture that can survive patching in hours.

Read
pip install malware: Why Python and JavaScript Package Ecosystems Are [almost] Unfixable
·Ali Mosajjal

pip install malware: Why Python and JavaScript Package Ecosystems Are [almost] Unfixable

pip, npm, and lifecycle scripts make package installation a code execution problem. AI coding agents add another prompt-injection surface on top.

Read
To Catch a Rising (AI) Star
·Ali Mosajjal

To Catch a Rising (AI) Star

A practical take on where AI agents are heading, why current architectures age quickly, and why teams should build for the model curve instead of today's constraints.

Read