AI Control

What OpenAI’s 20% Monitoring Overhead Should Change in Your Business Case In July 2026, during an internal cyber capability evaluation based on the ExploitGym benchmark, an autonomous agent system driven by a combination of OpenAI models escaped its sandbox. The evaluation ran without production cyber classifiers and with cyber refusals reduced. The models exploited a…

In August 2012, a Knight Capital Group algorithm executed nearly four million trades in 45 minutes, resulting in losses exceeding $440 million. The incident was not primarily a technology failure. It was a governance failure the consequence of automated behavior operating without adequate control. That lesson has not been fully absorbed. Today, AI and advanced…

A widely shared LinkedIn post from March 2026 explains, with commendable clarity, the distinction between Skills and MCP in agentic AI architecture. Skills teach the model how to reason. MCP gives the model access to external systems, CRMs, databases, internal documents, business tools. The post concludes: “Skills tell the model how to work. MCP lets…

What strikes you most in Davos isn’t the number of panels. It’s the concentration of people actually pushing the boundaries — from models and chips to energy systems, national security, and economic resilience. I’m leaving with a clear sense that we’re approaching a shift far larger than another wave of digitalization. Today, the real question…