Content
Strategic insights on blockchain, AI, security, leadership, and career in an age of disruption.
AI Agents and Access Controls: A New Risk
AI agents exhibit reward hacking behavior—cheating tests and deleting evidence—making least-privilege access controls critical before deploying them near financial close or client data.
Recent Essays
Stop Waiting for Better AI Models—Build Better Systems
Salesforce's research shows AI agents improved from 43.5% to 93% task completion without upgrading the model—by perfecting workflows, instructions, and evaluation. Your competitive edge isn't the AI itself; it's the system and scorecard you build around it.
Immutability's Hidden Cost: The Shutdown Problem
Blockchain's immutability prevents unauthorized ledger changes—but also prevents authorized shutdowns, leaving dormant permissions live and exploitable long after systems close.
The Real AI Risk Hiding in Your Company
78% of CISOs cite AI agent security as their top concern. The danger isn't the model—it's the unsanctioned agent with production access your team built last week.
Prediction Markets as Insurance: The Kalshi Hedge
A conference used prediction markets to hedge weather risk—turning a speculative exchange into an insurance desk. Here's why it works and what your CFO needs to know.
#AI supervision
1 post
