security | Journal

07·052026

Latest

CloudTrail Doesn't Have Your Prompts

I built an MCP server and discovered it didn't need to be an MCP. What a collapsed design taught me about CloudTrail, Bedrock invocation logs, and security tools that report clean when they cannot see.

AWS · BEDROCK · SECURITY · MCP · AI

02·202026

The Unintended Consequences of a Lying Machine

In 44 BC, Rome's most devoted republicans stabbed Caesar to stop a monarchy and got an empire. In 1902, French administrators in Hanoi offered a penny per rat tail and got rat farms. Both disasters share a cause. So does the way we are building AI.

STRATEGY · LIFE · AI · SECURITY

02·072026

Running Claude Code in Docker Sandboxes

DOCKER · CLAUDE-CODE · AI-AGENTS · SECURITY · DEVOPS

Working notes from the sharp end.

CloudTrail Doesn't Have Your Prompts

The Unintended Consequences of a Lying Machine

Running Claude Code in Docker Sandboxes